Commodity searching method and device, equipment and medium

Abstract

The invention relates to a commodity search method and device, equipment and a medium in the technical field of e-commerce search, and the method comprises the steps: obtaining a commodity search request triggered by a user, issuing the commodity search request to a search engine, analyzing the commodity search request by the search engine, and determining a corresponding search keyword and a target commodity search number; the search engine calls commodity indexes of corresponding quality grades one by one according to the search keyword and the target commodity search quantity and the quality grades of the commodity information from high to low; when a commodity search result which is related to the search keyword and meets the target commodity search quantity is searched from the called commodity index, calling is ended, and the commodity search result is returned; and according to the interest association score between each searched commodity and the user in the commodity search result, carrying out commodity sorting, constructing a corresponding commodity search list, and sending the commodity search list to the user. According to the method, the commodities interested by the user can be efficiently and accurately searched.

Description

Technical Field

[0001] The present application relates to the field of e-commerce search technology, and in particular to a product search method and its corresponding device, computer equipment, and computer-readable storage medium. Background Art

[0002] E-commerce platforms are the core places for e-commerce transactions and services, and the number of products on e-commerce platforms often reaches billions. When users are looking for specific products, they usually search by entering keywords, such as "red dress". E-commerce platforms will display a series of products based on the relevance of keywords. Users then browse these products to select the products they want. Therefore, the product search function is one of the indispensable core services of e-commerce platforms, which directly affects the progress of product transactions. Due to the huge number of products and search queries, the architectural design of product search is crucial to ensure platform performance and control service costs.

[0003] Product searches on e-commerce platforms implemented using traditional technologies usually search all products on all e-commerce platforms to obtain corresponding search results. The more products are searched, the worse the search performance becomes, the longer the search takes, and it can even lead to pain points such as search timeouts and unstable searches. Moreover, e-commerce platforms usually continue to add products, and the severity of these problems may gradually increase.

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

[0005] The primary purpose of the present application is to solve at least one of the above problems and to provide a product search method and its corresponding device, computer equipment, and computer-readable storage medium.

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

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

[0008] Obtain the product search request triggered by the user and send it to the search engine, which will parse the product search request and determine the corresponding search keywords and target product search quantity;

[0009] The search engine calls the product indexes of corresponding quality levels one by one according to the search keywords and the target product search quantity and in descending order of the quality level of the product information, until a product search result related to the search keywords and meeting the target product search quantity is found from the called product indexes, and the calling is terminated and the product search result is returned;

[0010] Determine the interest association score between each search product in the product search results and the user, sort the products according to the interest association score, construct a corresponding product search list and send it to the user.

[0011] On the other hand, a product search device provided to meet one of the purposes of the present application includes a request processing module, a result returning module and a request answering module, wherein the request processing module is used to obtain a product search request triggered by a user and send it to a search engine, and the search engine parses the product search request to determine the corresponding search keywords and the target product search quantity; the result returning module is used for the search engine to call the product indexes of corresponding quality levels one by one according to the search keywords and the target product search quantity, according to the quality level of the product information from high to low, until a product search result related to the search keyword and meeting the target product search quantity is searched from the called product index, the call is terminated and the product search result is returned; the request answering module is used to determine the interest association score between each search product in the product search result and the user, sort the products according to the interest association score, and construct a corresponding product search list to send to the user.

[0012] On the other hand, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the product search method described in the present application.

[0013] On the other hand, a computer-readable storage medium is provided to meet another purpose of the present application, which stores a computer program implemented according to the product search method in the form of computer-readable instructions, and when the computer program is called and executed by a computer, it executes the steps included in the method.

[0014] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0015] This application firstly sends the request to the background search engine when the user triggers a search request on the e-commerce platform, and the search engine parses the request to extract the search keywords entered by the user and the expected number of products to be obtained, that is, the target product search number.

[0016] Next, the search engine calls the corresponding product indexes in descending order of quality based on the search keywords and the number of target product searches. These indexes contain product information of different quality levels. The search engine will search these indexes one by one, looking for products related to the search keywords until a sufficient number of products are found to meet the user's search needs. In this process, the search engine will not only consider the relevance of the product to the search keywords, but also the quality level of the product, and give priority to searching and displaying those higher quality products to ensure that users can find high-quality products that they may be interested in more quickly, thereby improving the user experience.

[0017] Once the search engine has found a sufficient number of relevant products, it will stop further searching and start sorting the products based on the interest correlation scores between the products and the user. The search engine can personalize the order of the products so that those products that best suit the user's interests are at the front of the list.

[0018] Finally, the search engine will construct a product search list based on the sorting results and send it to the user. The list received by the user has been carefully pre-processed and sorted to provide the most relevant and most likely products to arouse the user's interest, significantly improve search efficiency, greatly reduce the search and display of relatively low-quality products, enhance the personalization of search results, and thus improve the user's shopping experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] 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:

[0020] Figure 1 The network architecture of the e-commerce platform exemplified in this application;

[0021] Figure 2 A flowchart of a typical embodiment of the commodity search method of the present application;

[0022] Figure 3 This is a functional block diagram of the product search device of this application;

[0023] Figure 4 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[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] In some embodiments, an application suitable for installation in a terminal device can be provided to serve the access needs of different users, so that various users can access the e-commerce platform 82 by running the application in the terminal device, such as the merchant backend module of the online store in the e-commerce platform 82, etc. In the process of implementing business activities through these functions, the e-commerce platform 82 can implement various functions related to supporting the implementation of business activities as middleware or online services and open corresponding interfaces, and then implant the toolkit corresponding to the interface access function into the application to implement function expansion and task implementation. The business management engine can include a series of basic functions, and expose these functions to online services and / or application calls through APIs. Online services and applications use corresponding functions by remotely calling corresponding APIs.

[0037] With the support of various components of the business management engine instance, the e-commerce platform 82 can provide online shopping functions, enabling merchants to establish connections with customers in a flexible and transparent manner. Consumer users can select items online, create product orders, provide the delivery address of the goods in the product order, and complete the payment confirmation of the product order. Then, the merchant can review and complete or cancel the order. The review component carried by the business management engine instance can realize the compliance use of business processes to ensure that the order is suitable for fulfillment before actual fulfillment. Orders may sometimes be fraudulent and need to be verified (such as ID card checks). There is a payment method that requires merchants to wait to ensure that funds are received, which can play a role in preventing such risks, etc. Order risks may be generated by fraud detection tools submitted by third parties through order risk APIs, etc. Before fulfillment, merchants may need to obtain payment information or wait to receive payment information in order to mark the order as paid, so that the merchant can prepare to deliver the product. Such situations can be reviewed accordingly. The review process can be implemented by the fulfillment component. Merchants can use the fulfillment component to review, adjust work, and trigger related fulfillment services, such as: manual fulfillment services, which are used when merchants pick and pack products in boxes, purchase shipping labels and enter their tracking numbers, or simply mark items as fulfilled; custom fulfillment services, which can define email notifications; API fulfillment services, which can trigger third-party applications to create fulfillment records in third parties; legacy fulfillment services, which can trigger custom API calls from the business management engine to third parties; gift card fulfillment services. You can provide the generation of numbers and activate gift cards. Merchants can use the order printer application to print shipping orders. The fulfillment process can be executed when the items are packed in boxes and ready for shipping, tracking, delivery, consumer receipt verification, etc.

[0038] It can be seen that the services provided by the e-commerce platform are based on products. The corresponding commodity data is the basic data of the e-commerce platform. The commodity information is provided through the commodity data. The mining and utilization of commodity data is the basis for realizing various technical services, including the use of user transaction data and commodity data in the commodity data of the e-commerce platform to provide basic services for the operation of the real-time streaming data processing system. Therefore, the real-time streaming data processing system of this application can be run in any one or more servers of the cluster of the e-commerce platform, so as to realize various functions by using various commodity data provided by the e-commerce platform.

[0039] A product search method of the present application can be programmed as a computer program product and deployed in a client or server for execution. For example, in the exemplary application scenario of the present application, it can be deployed and implemented in a server of an e-commerce customer service platform, thereby accessing an interface opened after the computer program product is running and performing human-computer interaction with the process of the computer program product through a graphical user interface to execute the method.

[0040] See also Figure 2 The commodity search method of the present application, in its typical embodiment, comprises the following steps:

[0041] Step S1100: Obtain the product search request triggered by the user and send it to the search engine, which parses the product search request to determine the corresponding search keywords and the number of target product searches;

[0042] The search engine is implemented using Elasticsearch, which is a full-text search engine based on Lucene, also known as a distributed search engine, developed in Java, and open source, with distribution of Http Web and frameless Json documents.

[0043] The user can enter the search keyword in the search box provided, and then encapsulate the default product search quantity as the target product search quantity and the search keyword to obtain the corresponding product search request and send it to the server. The default product search quantity can be preset by those skilled in the art according to business needs.

[0044] The server receives the commodity search request and responds to it, forwarding the request to any node in the search engine cluster, and using the node as a coordination node, the coordination node parses the request, determines the search keyword, the number of target commodity searches, and a preset index routing rule. The index routing rule stipulates that the coordination node preferentially searches for commodity indexes with relatively high quality levels, while commodity indexes with relatively low quality levels are searched after the previous search results are insufficient.

[0045] Step S1200: the search engine calls product indexes of corresponding quality levels one by one according to the search keyword and the target product search quantity, in descending order of the quality level of product information, until a product search result related to the search keyword and meeting the target product search quantity is found from the called product indexes, the call is terminated, and the product search result is returned;

[0046] The coordination node in the search engine calls the commodity indexes of the corresponding quality levels one by one according to the index routing rules and the quality level of the commodity information from high to low, that is, the commodity index with the highest quality level is called first, and the nodes of the shards storing the commodity index in the cluster are determined. Then, the coordination node broadcasts the commodity search request to the nodes, and uses these nodes as data nodes. The local data nodes of each shard perform searches according to the search keywords in the commodity search request, and build a priority queue of the size of the whole set. Then, the data nodes where each shard is located return all commodity identifiers in their respective priority queues, and perform any one or multiple data operations such as data merging, sorting, and paging on all commodity identifiers according to the search association score of each commodity identifier, and output a commodity identifier set. The coordination node accesses the data node that submits the commodity identifier according to the commodity identifier in the commodity identifier set, and reads the commodity information of the corresponding commodity identifier from the data node to form a search result. The commodity identifier is used to uniquely refer to the corresponding commodity, which is distinguished from other commodities. The search relevance score can be implemented as needed by those skilled in the art, for example, based on the rule matching degree, semantic matching degree, etc. between the searched product information and the search keywords. The rule matching degree can be the category relevance degree, whether the product title is hit, etc., and the semantic matching degree can be the semantic relevance degree between the text information in the product information and the search keywords. It can be specifically implemented using a deep learning algorithm suitable for the NLP field.

[0047] Determine whether the total number of search products included in the search results is less than the target product search quantity. If so, the coordination node continues to search for the product index of the quality level next to the most recently searched quality level, obtains the corresponding search result and appends it after the last search product in the most recently obtained search result, and repeats the determination. If greater than or equal to, the coordination node terminates the search and returns the most recently obtained search result to the server.

[0048] Step S1300: determine the interest association score between each search product in the product search results and the user, sort the products according to the interest association score, construct a corresponding product search list and send it to the user.

[0049] Call a preset interest assessment model, which is pre-trained to a convergence state, and learns the ability to predict interest association scores including click-through rate, conversion rate, and click-to-conversion rate between corresponding users and products based on input user feature information and product feature information. Those skilled in the art can flexibly implement the interest assessment model, and can also refer to the further disclosure of some subsequent embodiments for implementation. This step will not be discussed for the time being.

[0050] In one embodiment, an interest evaluation model based on the ESMM model is used to obtain the user characteristic information of the user, and for each search product in each product search result, the product characteristic information of the search product is obtained, and the interest evaluation model determines the click rate, conversion rate, and click-to-conversion rate between the user and the search product as the interest association score based on the product characteristic information and the user characteristic information. Thus, the interest association score of each search product can be obtained. Further, for each interest association score, the click rate, conversion rate, and click-to-conversion rate in the interest association score are multiplied by their respective weights and then added to obtain a ranking score. All search products are sorted in order from high to low according to the ranking score, and multiple search products with the highest ranking are screened out or all search products after sorting are directly added to the product search list that is initially constructed as empty. After that, the product search list is sent to the user to respond to the product search request.

[0051] It is not difficult to understand from the above embodiments that, compared with the prior art, the present application has many advantages, including at least:

[0052] This application firstly sends the request to the background search engine when the user triggers a search request on the e-commerce platform, and the search engine parses the request to extract the search keywords entered by the user and the expected number of products to be obtained, that is, the target product search number.

[0053] Next, the search engine calls the corresponding product indexes in descending order of quality based on the search keywords and the number of target product searches. These indexes contain product information of different quality levels. The search engine will search these indexes one by one, looking for products related to the search keywords until a sufficient number of products are found to meet the user's search needs. In this process, the search engine will not only consider the relevance of the product to the search keywords, but also the quality level of the product, and give priority to searching and displaying those higher quality products to ensure that users can find high-quality products that they may be interested in more quickly, thereby improving the user experience.

[0054] Once the search engine has found a sufficient number of relevant products, it will stop further searching and start sorting the products based on the interest correlation scores between the products and the user. The search engine can personalize the order of the products so that those products that best suit the user's interests are at the front of the list.

[0055] Finally, the search engine will construct a product search list based on the sorting results and send it to the user. The list received by the user is carefully screened and sorted to provide the most relevant and most likely products to arouse the user's interest, significantly improve search efficiency, significantly reduce the search and display of relatively low-quality products, enhance the personalization of search results, and thus improve the user's shopping experience.

[0056] In a further embodiment, step S1200, the search engine calls the product indexes of corresponding quality levels one by one according to the search keywords and the target product search quantity, according to the quality level of the product information from high to low, until a product search result related to the search keywords and meeting the target product search quantity is found from the called product indexes, before the call is terminated and the product search result is returned, includes the following steps:

[0057] Step S2200: Obtain the latest product information of each product on the e-commerce platform;

[0058] To ensure the accuracy of product information, the product quality score of each product on the e-commerce platform is established based on the latest product information of the product. That is to say, whenever the product information of any product is updated to the latest, the product quality score of the product is determined by the latest product information.

[0059] Step S2210: for each product information, determine a corresponding basic experience score according to the degree to which the product information satisfies the basic experience of the user, and determine a corresponding positive optimization score according to the degree to which the product information improves the basic experience of the user, and determine a product quality score for the product information based on the basic experience score and the positive optimization score;

[0060] The completeness, accuracy and compliance of the text information in the product information can be evaluated, and the clarity and compliance of the image information in the product information can be evaluated to obtain quantitative scores corresponding to the two aspects of the evaluation, and then the two quantitative scores can be weighted and fused to obtain the basic experience score. The completeness of the text information can be whether the text information fully describes the necessary information of the corresponding product; the accuracy can be whether the text information contains false propaganda and / or falsely marked excessive prices and / or typos and / or missing words; the compliance can be whether the text information violates the platform transaction rules and / or violates legal requirements and / or false propaganda and / or deliberately piles up words to improve retrieval relevance. The clarity of the image information can be the clarity of the image information showing the corresponding product; the compliance can be whether the image information violates the platform transaction rules and / or violates legal requirements.

[0061] The richness and / or selling point expression of the product information can be evaluated. It can be understood that the more selling points of the corresponding product are described and displayed, the more display perspectives, the more marketing elements, etc., any one or any combination thereof, the higher the richness of the product is evaluated to be; the more prominent, novel, creative, etc., the more selling points of the corresponding product are described and displayed, the higher the selling point expression of the product is evaluated to be. Thus, the quantitative score corresponding to the richness and / or selling point expression is evaluated, and any one of the two quantitative scores, or the result value of the weighted fusion of the two quantitative scores, is used as the positive optimization score of the product information.

[0062] The above-mentioned evaluations can be performed through manual assessment, rule judgment, deep learning algorithm, etc., and those skilled in the art can implement them flexibly.

[0063] The basic experience score and the positive optimization score are multiplied by their respective preset weights and then added together to obtain the product quality score. The preset weights of the basic experience score and the preset weights of the positive optimization score can be pre-set by technical personnel in this field according to business requirements, for example, to 0.6 and 0.4 respectively.

[0064] Step S2220, obtaining the grade scoring intervals associated with each quality grade divided from high to low, and determining the quality grade to which each product information belongs according to the grade scoring interval to which the product quality score of each product information belongs;

[0065] The various quality levels and their associated grade scoring intervals may be pre-set by those skilled in the art as needed. For example, the quality levels: high, medium, low, and the corresponding grade scoring intervals: [1, 0.85), [0.85, 0.35], (0.35, 0].

[0066] First, the grade scoring interval to which the product quality score of each product information belongs is determined, and then the quality grade to which each product information belongs is confirmed as the grade scoring interval to which the product quality score of the product information belongs.

[0067] Step S2230: For each quality level, construct a commodity index of the quality level according to all commodity information belonging to the quality level.

[0068] For each quality level, a commodity index of the quality level is created in the search engine, and all commodity information of the quality level is written into the commodity index, so that each commodity information is equivalent to a document stored in the commodity index.

[0069] In this embodiment, by obtaining the latest information of the product and establishing a product quality score based on the information, the timeliness and accuracy of the product information are ensured, and the two aspects of product information satisfaction and improvement of the user's basic experience are evaluated. The contribution of product information to the user experience can be comprehensively measured, and the product information is classified into different quality levels. A corresponding product index is constructed for each quality level, so that the search engine can subsequently call the product index of a relatively high quality level according to the user's search needs, thereby quickly retrieving products that are related to the search keywords and meet the user's needs, which not only improves the search efficiency, but also ensures the relevance and quality of the search results, so that users can find the high-quality products they are interested in more quickly.

[0070] In a further embodiment, before step S1300, determining the interest association score between each search product in the product search results and the user, the following steps are included:

[0071] Step S2300: obtaining training samples and supervisory labels in a prepared training set, wherein the training samples include key product information of a product and key user information of a user, and the supervisory labels represent whether the user has shown interest in the product;

[0072] The key information of the product includes all pre-processed texts describing the product, i.e., text information, and / or statistical features of the product. The text describing the product may be any one or more of the product category, product label, product title, product attributes, etc.; the statistical features may be the cumulative number of any one or more of the historical exposure, clicks, purchases, etc. of the product.

[0073] The user key information includes the pre-processed user portrait, and / or the description text of the product when the user shows interest in the product, which can be any one or more of the product category, product label, product title, product attributes, etc.; the user portrait includes the user's basic information and / or user behavior data, and the user's basic information usually includes any one or more of gender, age, occupation, purchasing power, geographic location, product brand preference, product category preference, product price preference, etc.; the user behavior data includes statistics on specified interactive behaviors between users and products and / or independent station stores, such as the time period and / or frequency of users visiting independent station stores, and the frequency and / or time period of users browsing products, searching for products, adding to shopping carts, purchasing products, clicking on products, collecting products, and commenting on products in independent station stores.

[0074] The interest behavior may be any one or a combination of click, add to cart, purchase, and repeat purchase.

[0075] The preprocessing is to format the processed information into a format suitable for model processing, which can be any one or more of one-hot encoding, feature normalization, regularization, etc., and those skilled in the art can flexibly implement it.

[0076] As for the training samples and their supervisory labels prepared in advance in the training set, specifically, by monitoring the interaction between users and products in the independent station store, when the user shows interest in the product, the user key information of the user and the product key information of the product are collected as training samples, and the supervisory label of the training sample is marked as representing that the user has shown interest in the product, which can be any one or any multiple of the click-through rate (CTR) of 1, the conversion rate (CVR) of 1, the click-through conversion rate (CTCVR) of 1, etc.; when the user does not show interest in the product, the user key information of the user and the product key information of the product are collected as training samples, and the supervisory label of the training sample is marked as representing that the user does not show interest in the product, which can be any one or any multiple of the click-through rate (CTR) of 0, the conversion rate (CVR) of 0, the click-through conversion rate (CTCVR) of 0, etc.

[0077] Step S2310, inputting the training sample into the interest evaluation model to extract the corresponding deep semantic features and obtain the interest feature vector;

[0078] It is recommended to use ESMM in the selection of the interest assessment model. Other variant models such as LR, DeepFM model, etc. that can achieve the same functions of this application can also be replaced by it.

[0079] In one embodiment, an interest assessment model based on ESSM is adopted, and each supervision label in the training set includes the click rate, click conversion rate, and conversion rate corresponding to the corresponding training sample. The training samples are input into the interest evaluation model, and the shared embedding layer in the interest evaluation model maps each high-dimensional sparse feature in the training samples to a corresponding low-dimensional dense feature. Furthermore, the first branch is that the pooling layer in the click prediction tower in the interest evaluation model performs bitwise element addition operation on each low-dimensional dense feature on the user side to obtain a user comprehensive feature, and at the same time, performs bitwise element addition operation on each low-dimensional dense feature on the product side to obtain a product comprehensive feature, and then horizontally splices the user comprehensive feature and the product comprehensive feature to obtain a deep semantic feature representing the possibility of clicking, and its corresponding vectorized representation, i.e., a behavioral feature vector for predicting the click-through rate; the second branch is that the pooling layer in the conversion prediction tower in the interest evaluation model performs bitwise element addition operation on each low-dimensional dense feature on the user side to obtain a user comprehensive feature, and at the same time, performs bitwise element addition operation on each low-dimensional dense feature on the product side to obtain a product comprehensive feature, and then horizontally splices the user comprehensive feature and the product comprehensive feature to obtain a deep semantic feature representing the possibility of conversion, and its corresponding vectorized representation, i.e., a behavioral feature vector for predicting the conversion rate.

[0080] Step S2320: classify and map the interest feature vector to predict the corresponding interest association score;

[0081] Following the first branch, the multi-layer perceptron (MLP) in the click prediction tower in the interest evaluation model performs a nonlinear transformation on the behavioral feature vector used to predict the click-through rate, and then activates the sigmoid or softmax function to obtain the predicted click-through rate; following the second branch, the multi-layer perceptron (MLP) in the conversion prediction tower in the interest evaluation model performs a nonlinear transformation on the behavioral feature vector used to predict the conversion rate, and then activates the sigmoid or softmax function to obtain the predicted conversion rate. In addition, the predicted click-through rate and the predicted conversion rate are multiplied by the interest evaluation model to obtain the predicted click-through conversion rate. Therefore, the predicted click-through rate, the predicted conversion rate, and the predicted click-through conversion rate are the interest association scores corresponding to the training samples.

[0082] Step S2330, calculate the loss value corresponding to the predicted interest association score based on the supervisory label, and update the interest assessment model based on the loss value until the model converges, so as to determine the interest association score between each search product in the product search results and the user.

[0083] Call the preset cross entropy loss function, and calculate the cross entropy loss values ​​corresponding to the predicted click-through rate, predicted conversion rate, and predicted click-to-purchase rate according to the supervisory labels of the training samples. When the cross entropy loss values ​​reach the preset threshold, it indicates that the interest evaluation model has been trained to a convergence state, so that the interest evaluation model training can be terminated; when any cross entropy loss value does not reach the preset threshold, it indicates that the interest evaluation model has not converged, so the model is gradient updated according to the cross entropy loss value, usually by back propagation to correct the weight parameters of each link of the model so that the model is closer to convergence, and then continue to call other training samples and their supervisory labels to perform iterative training on the interest evaluation model until the model is trained to a convergence state. The preset threshold can be pre-set by a person skilled in the art as needed according to the disclosure herein.

[0084] In this embodiment, the interest assessment model is trained until it reaches a convergence state, so that it can learn the ability to reliably and accurately predict any one or more of the interest association scores including the click-through rate, conversion rate, and click-to-conversion rate between the corresponding user and the product based on the input user feature information and product feature information.

[0085] In a further embodiment, after step S2230, for each quality level, constructing a commodity index of the quality level according to all commodity information belonging to the quality level, the following steps are included:

[0086] Step S2231: Determine various performance levels divided from high to low and the performance level to which each node belongs according to the search service performance of each node in the search engine cluster;

[0087] Elasticsearch is a distributed search engine built on Lucene that allows users to interact through an HTTP Web interface and JSON document format. In Elasticsearch, a cluster is a network of multiple nodes that work together to provide high availability and scalability. Each node is part of a cluster, and they work together to share data and load to achieve functions such as full-text search, analysis, and storage. The nodes coordinate task allocation and data synchronization through internal communication, allowing the entire cluster to operate like a single search engine, providing fast and consistent search results. In short, an Elasticsearch cluster is composed of multiple nodes that work together to form a powerful, scalable search and analysis platform.

[0088] The search service performance of the node may depend on the hardware performance index and / or computing performance index of the node, wherein the hardware performance index includes any one or more of CPU processing capability, GPU processing capability, memory read / write speed, storage read / write speed, network transmission efficiency, etc. The computing performance index includes any one or more of QPS (query rate per second), response time, throughput, etc.

[0089] In one embodiment, a performance evaluation algorithm can be constructed, which will comprehensively consider the hardware performance index and / or the operation performance index to score each node. Specifically, it can be implemented based on weighted fusion scoring, in which different performance indicators are assigned different weights according to the degree of influence of the performance indicator on the search service, so that for a single node, each performance indicator is multiplied by its respective weight to obtain the search service performance score of the node. Further, all nodes are sorted in order from high to low according to the search service performance score to obtain the corresponding node sequence, and according to the total ratio of each performance level divided from high to low, the total ratio of each performance level is multiplied by the total number of nodes in the node sequence to determine the total number of nodes corresponding to each performance level. On this basis, according to the order of performance levels from high to low, the ranking interval corresponding to the performance level is determined one by one, thereby, each node in the node sequence can determine the performance level corresponding to the ranking interval to which the node belongs according to its ranking. The sum of the total ratios of each performance level is 1, and each specific total ratio can be set as needed by a technician in this field. The total number of the divided performance levels is equal to the total number of the quality levels.

[0090] Step S2232: For each performance level, the commodity index of the quality level matching the performance level is stored in fragments in the nodes belonging to the performance level, so that nodes with higher performance levels store fragments of commodity indexes with higher quality levels.

[0091] It can be understood that both performance levels and quality levels have high and low levels, and the total number of the two is the same. Therefore, the performance level and quality level of the same level can be confirmed to match.

[0092] By setting the total number of primary shards corresponding to commodity indexes of each quality level in the configuration environment of the search engine, and setting each commodity index to be distributedly stored by all nodes belonging to a performance level matching the quality level of the commodity index, each index shard can be stored in the corresponding node. The above settings can be set as needed by those skilled in the art.

[0093] In this embodiment, by comprehensively evaluating the hardware and computing performance indicators of each node, the nodes can be accurately divided into different performance levels, ensuring that the performance of each node can be fully utilized later, so that commodity indexes of different quality levels can be allocated to nodes of corresponding performance levels, thereby ensuring that high-level performance nodes store high-level quality commodity index shards, and low-level performance nodes store low-level quality commodity index shards. This not only improves the efficiency of the search service, but also optimizes resource allocation, so that high-performance nodes can efficiently process more complex or more frequent queries, while low-performance nodes can promptly process relatively simple tasks, providing users with a faster and more accurate search experience.

[0094] In a further embodiment, after step S1300, constructing a corresponding product search list and sending it to the user, the following steps are included:

[0095] Step S1400: acquiring the searched products in the product search list that the user has shown interest in as interest searched products;

[0096] The front end can monitor the user's interactive behaviors with respect to the searched products in the product search list, and the interactive behaviors include all behaviors with respect to the searched products in the product search list, such as scrolling, sharing, liking, collecting, clicking, adding to a shopping cart, purchasing, repurchasing, finding similar products, setting filtering conditions, etc. The interest behaviors include any one or any combination of clicking, adding to a shopping cart, purchasing, repurchasing, finding similar products, and setting filtering conditions. Thus, the searched products for which the user performs any one or any combination of clicking, adding to a shopping cart, purchasing, repurchasing, and finding similar products in the interest behaviors, and after the user performs the setting filtering conditions in the interest behaviors, the remaining searched products are filtered out as the interest searched products.

[0097] Step S1410: extracting the user's interest description text according to the product information of each interest search product;

[0098] In one embodiment, based on the importance rule of e-commerce entity types, the importance rule divides different entity types into three levels: high, medium, and low. For example, "brand, category" is in the high-end, which is the most important; secondly, "style, design, color, season, crowd, location..." is in the medium-end; and finally, "size, modifier, service impact, series, unit..." is in the low-end. The specific division levels and entity types belonging to each level can also be flexibly implemented by those skilled in the art. Thus, the text information in the product information of each of the interest search products is extracted, and the text corresponding to the high-end and medium-end entity types is determined as the interest text. After deduplication of these interest texts, they are aggregated to form the interest description text of the user.

[0099] Step S1420: update the search keyword according to the interest description text, obtain a target search keyword and send it to the user;

[0100] The search keywords are updated using a large language model. Specifically, a prompt template is preset, and the prompt template includes a task description, an interest description text to be embedded, and a search keyword. A person skilled in the art may refer to the following disclosure to flexibly set the prompt template. The task description in the prompt template is exemplified as follows: "Please extract the most relevant keywords from the interest description text based on the following customer's interest description text and the previous search keywords, so that new search keywords are generated on the basis of the search keywords. These keywords should be able to accurately reflect the customer's deeper interests or preferences on the basis of the original ones, and help optimize the search results to make them more personalized and accurate. Interest description text: {$interest description text to be embedded$}, search keywords: {$search keywords to be embedded$}." The interest description text and the search keywords are embedded in the prompt template, and the corresponding prompt text is obtained and input into the large language model, and the large language model outputs the corresponding new search keywords as target search keywords, which are sent to the user for the user to review the target search keywords, and the user is informed whether to confirm the product search with the target search keywords.

[0101] Step S1430: In response to the user confirming the target search keyword, a corresponding product search request is triggered, and a product search list that responds to the product search request is constructed and sent to the user.

[0102] When the user confirms the target search keyword, it usually means that the user confirms that the target search keyword matches his or her actual search intention, or that the search intention further stated by the target search keyword is what the user is interested in. This triggers the construction of a product search request that encapsulates the default product search quantity and the target search keyword and sends it to the server.

[0103] The server receives the product search request and responds to it, and according to the relevant steps from step S1100 to step S1400, it calls the search engine accordingly to determine the product search list required by the product search request and sends it to the user.

[0104] In this embodiment, the user's interest behavior is identified by monitoring the user's interactive behavior in the product search list to capture the user's interest products. The interest description text is extracted according to the user's interest products. In this process, the importance of different entity types in the product information is evaluated and classified to ensure that the extracted information is comprehensive and representative, which can not only extract the user's core interest points, but also reduce the redundancy of information to a certain extent, making the interest description text more refined and accurate. Subsequently, the interest description text is analyzed using a large language model to update the user's previous search keywords, which can transform the user's implicit interest into an explicit search intention, thereby optimizing the search results to make them more in line with the user's personalized needs, further improve the accuracy and personalization level of the search, and help users find the products they are really interested in faster. Finally, when the user confirms the target search keyword, a corresponding product search request is constructed, and the corresponding product search list is returned in response to the request, so as to ensure that the user can immediately obtain the matching product information after clarifying his search intention, thereby improving the user's search efficiency and satisfaction.

[0105] To sum up, providing users with a more accurate, personalized and responsive search experience can not only improve users' shopping satisfaction, but also enhance the user stickiness and conversion rate of e-commerce platforms.

[0106] In a further embodiment, step S1410, extracting the user's interest description text according to the product information of each interest search product, includes the following steps:

[0107] Step S1411, searching for product information of products according to the respective interests, and extracting a product commonality information set and a product characteristic information set;

[0108] In one embodiment, the text information in the product information of each of the interest search products is segmented to obtain a segmentation sequence corresponding to each of the text information, and the same word units in each segmentation sequence are extracted and added to the product commonality information set that is initially created as empty. The segmentation can be implemented based on an e-commerce dictionary and named entity recognition.

[0109] In addition, it is understood that the TF-IDF algorithm or the BM25 algorithm can be used in advance to determine the product characteristic information in the text information of the product information of each product in the e-commerce platform. Thus, the product characteristic information in the product information of each of the interest search products is added to the product characteristic information set initialized to be empty.

[0110] Step S1412: using a large language model to describe the user's interests according to the commodity commonality information set and the commodity characteristic information set, so as to obtain a corresponding interest description text.

[0111] A large language model is used to describe the user's interests. Specifically, a prompt template is preset, and the prompt template includes a task description and an information set to be embedded. Those skilled in the art may refer to the following disclosure to flexibly set the prompt template. The task description in the prompt template is exemplified as follows: "Please describe the customer's interests based on the common information and characteristic information of the products in the product search results clicked by the customer given below, and output the corresponding interest description text. Common information: {$common information set of products to be embedded$}; characteristic information: {$characteristic information set of products to be embedded$}." The large language model and the prompt template are called, and the common information set and the characteristic information set of the products are embedded into the common information set and the characteristic information set of the products to be embedded in the prompt template to obtain the corresponding prompt text, and the prompt text is input into the large language model to obtain the interest description text output by the large language model.

[0112] In this embodiment, firstly, by extracting the common information set and the characteristic information set of the commodities, the user's interests can be captured more comprehensively. Since the common information set reflects the characteristics of the user's common or repeated attention in multiple commodities, and the characteristic information set highlights the user's unique preference in a specific commodity, the generated interest description text can not only comprehensively interpret the user's interests, but also deeply explore the user's personalized needs. Furthermore, a large language model is used to describe the user's interests to obtain the interest description text, which can be used to update the original search keywords and further recommend the users the commodities they actually want.

[0113] In a further embodiment, step S2210, for each product information, determining a corresponding basic experience score according to the degree to which the product information satisfies the basic experience of the user, and determining a corresponding positive optimization score according to the degree to which the product information improves the basic experience of the user, and determining a product quality score of the product information based on the basic experience score and the positive optimization score, includes the following steps:

[0114] Step S2211: using a preset dual-tower model, the text identification branch thereof identifies the completeness, accuracy and compliance of the text information in the product information to obtain a corresponding text experience score;

[0115] The dual-tower model includes a text identification branch, an image identification branch, and a linear layer connected to the two branches; the text identification branch includes a text feature extraction layer and a classifier connected thereto, the text feature extraction layer is suitable for extracting the semantics of the input text for vector representation, and can be selected from a variety of known models, including but not limited to Bert, RNN, BiLSTM, BiGRU, RoBERTa, ALBert, ERNIE, BERT-WWM, etc., the classifier is suitable for multi-classification tasks, and can be MLP (feedforward neural network) or FC (fully connected layer); the image identification branch includes an image feature extraction layer and a classifier connected thereto, the image feature extraction layer is suitable for extracting the semantics of the input image for vector representation, and can be selected from a variety of known models, including but not limited to ViT, CNN model, deep convolution model VGG, Resnet, etc. The classifier is suitable for multi-classification tasks, and can be MLP (feedforward neural network) or FC (fully connected layer).

[0116] The deep semantic information corresponding to the text information in the commodity information is extracted by the text feature extraction layer in the text identification branch in the twin-tower model, and the deep semantic information includes the semantic and grammatical information of the text information. Then, the deep semantic information is vectorized to obtain a text feature vector, and the text feature vector is subjected to multi-classification mapping by a classifier connected to the text feature extraction layer, including: mapping to binary categories representing the completeness of various types of texts describing the corresponding commodity in the text information, and obtaining the classification probability corresponding to the completeness as the completeness, wherein the various types of texts include any multiple items of different commodity attributes, commodity categories, commodity titles, commodity details texts, etc.;

[0117] Map to binary categories representing the accuracy of various types of texts describing corresponding products in text information, such as the possibility of falsely marking excessive product prices, the possibility of false advertising, whether there are typos and / or missing words, etc., and obtain the classification probability corresponding to the accuracy as the accuracy; map to binary categories representing the compliance of various types of texts describing corresponding products in text information, such as the possibility of intentionally stacking words to improve search relevance, the possibility of violating platform transaction rules, the possibility of violating legal requirements, etc., and obtain the classification probability corresponding to the compliance as the compliance. The obtained completeness, accuracy and compliance are used as the text experience score.

[0118] Step S2212: the picture identification branch in the dual-tower model identifies the clarity and compliance of the picture information in the product information to obtain a corresponding picture experience score;

[0119] The deep semantic information corresponding to the image information in the product information is extracted by the image feature extraction layer in the image identification branch in the double tower model. The deep semantic information includes basic image features such as the edge, contour, color, texture, shape, etc. of the image information and high-level logical features of the clarity of the product display. Then, the deep semantic information is vectorized to obtain an image feature vector, and the classifier connected to the image feature extraction layer performs multi-classification mapping on the image feature vector, including: mapping to a binary category representing the clarity of each product image in the image information, and obtaining a classification probability corresponding to the clarity as clarity; mapping to a binary category representing the compliance of each product image in the image information, such as any one or more of the possibility of violating the platform transaction rules, the possibility of violating the legal requirements, etc., and obtaining a classification probability corresponding to the compliance as compliance. The obtained clarity and compliance are used as the image experience score.

[0120] Step S2213: linearly fuse the text experience score and the picture experience score by the linear layer in the dual-tower model to obtain a corresponding basic experience score;

[0121] The linear layer in the dual-tower model is connected to the text identification branch and the image identification branch, and the text experience score and the image experience score corresponding to the output of these two branches are received. The completeness, accuracy and compliance in the text experience score, and the clarity and compliance in the image experience score are multiplied by their respective weights and then added together to obtain the basic experience score. It can be understood that the five weights are learned by the linear layer in the dual-tower model during the training process.

[0122] According to the above disclosure of the reasoning process of the twin-tower model, those skilled in the art can flexibly and adapt the twin-tower model to be trained to a convergence state in advance in preparation for the above steps.

[0123] Step S2214: using a preset product evaluation model to evaluate the richness of the product information and the performance of the selling points, and obtaining a corresponding positive optimization score;

[0124] The product evaluation model includes an image representation layer and a text representation layer, a picture and text fusion layer connected to the picture representation layer and the text representation layer, and a classifier connected to the picture and text fusion layer. The recommended model for the picture representation layer is the Resnet model, and any other model such as EfficientNet, DenseNet, MobileNet, LeNet, ViT, VGG, etc. can also be used. The recommended model for the text representation layer is Bert, and any other model such as T5, BiLSTM, etc. can be used. The picture and text fusion layer can be implemented by weighted averaging, gating, etc., and those skilled in the art can choose one of them as needed. The classifier is suitable for multi-classification tasks and can be an MLP (feedforward neural network) or FC (fully connected layer).

[0125] The deep semantic information of the image information in the product information is extracted by the image representation layer in the product evaluation model, and the image semantic representation of the deep semantic information is obtained. The deep semantic information of the text information in the product information is extracted by the text representation layer in the product evaluation model, and the text semantic representation of the deep semantic information is obtained. The image and text fusion layer in the product evaluation model fuses the image semantic representation with the text semantic representation and maps them to the multimodal semantic space to obtain the corresponding image and text fusion representation. The classifier in the product evaluation model performs multi-classification mapping on the image and text fusion representation, including: a binary category representing the richness of describing and displaying the corresponding product, and obtaining the classification probability corresponding to the richness as the richness; a binary category representing the expression degree of the selling point for describing and displaying the corresponding product, and obtaining the classification probability corresponding to the expression degree as the selling point expression degree. The obtained richness and selling point expression degree are used as positive optimization scores.

[0126] It can be understood that the more selling points of the corresponding product are described and displayed, the more display perspectives, the more marketing elements, etc., any one or any combination of the above, the higher the richness of the product will be evaluated; the more prominent, novel, and creative the selling points of the corresponding product are described and displayed, the higher the selling point expression of the product will be evaluated.

[0127] According to the above disclosure of the reasoning process of the product evaluation model, those skilled in the art can flexibly and adapt the method to pre-train the product evaluation model to a convergence state in preparation for the above steps.

[0128] Step S2215: Perform weighted fusion on the basic experience score and the forward optimization score to obtain a corresponding product quality score.

[0129] The basic experience score and the positive optimization score are multiplied by their respective preset weights and then added together to obtain the product quality score. The preset weights of the basic experience score and the preset weights of the positive optimization score can be pre-set by technical personnel in this field according to business requirements.

[0130] In this embodiment, the text and pictures in the product information are deeply analyzed by using the twin tower model, and their corresponding text experience scores and picture experience scores are evaluated respectively, so as to obtain the basic experience score. Then, the richness of the product information and the performance of the selling points are further evaluated by the product evaluation model to obtain the positive optimization score. Finally, the basic experience score and the positive optimization score are weighted and fused to obtain the product quality score, which can comprehensively consider all aspects of the product information, including not only the completeness, accuracy, compliance, clarity, etc. of the product information, but also the attractiveness and marketing effect of the product information, so that the impact of the product information on the user experience can be more accurately reflected.

[0131] See also Figure 3 , a product search device provided to meet one of the purposes of the present application is a functional embodiment of the product search method of the present application. On the other hand, the device, a product search device provided to meet one of the purposes of the present application includes a request processing module 1100, a result return module 1200 and a request response module 1300, wherein the request processing module 1100 is used to obtain the product search request triggered by the user and send it to the search engine, and the search engine parses the product search request to determine the corresponding search keywords and the target product search quantity; the result return module 1200 is used for the search engine to call the product indexes of the corresponding quality levels one by one according to the search keywords and the target product search quantity, according to the quality level of the product information from high to low, until the product search results related to the search keywords and meeting the target product search quantity are searched from the called product index, the call is terminated and the product search results are returned; the request response module 1300 is used to determine the interest association score between each search product in the product search results and the user, sort the products according to the interest association score, and construct a corresponding product search list and send it to the user.

[0132] In a further embodiment, before the result return module 1200, it includes: an information acquisition submodule, which is used to obtain the latest product information of each product in the e-commerce platform; a score determination submodule, which is used to determine, for each product information, a corresponding basic experience score according to the degree to which the product information satisfies the user's basic experience, and determine the corresponding positive optimization score according to the degree to which the product information improves the user's basic experience, and determine the product quality score of the product information based on the basic experience score and the positive optimization score; a grade determination submodule, which is used to obtain the grade score intervals associated with each quality grade divided from high to low, and determine the quality grade to which each product information belongs according to the grade score interval to which the product quality score of each product information belongs; an index construction submodule, which is used to construct a product index of the quality grade according to all product information belonging to the quality grade for each quality grade.

[0133] In a further embodiment, before the request response module 1300, it includes: a training preparation submodule, which is used to obtain training samples and their supervisory labels in a prepared training set, wherein the training samples include the product key information of the product and the user key information of the user, and the supervisory label represents whether the user has shown interest in the product; a vector representation submodule, which is used to input the training samples into the interest evaluation model to extract the corresponding deep semantic features and obtain the interest feature vector; a score prediction submodule, which is used to classify and map the interest feature vector and predict the corresponding interest association score; a model convergence submodule, which is used to calculate the loss value corresponding to the predicted interest association score according to the supervisory label, and update the interest evaluation model according to the loss value until the model converges, so as to determine the interest association score between each search product in the product search results and the user.

[0134] In a further embodiment, after the index construction submodule, it includes: a level division and attribution submodule, which is used to determine the various performance levels divided from high to low and the performance level to which each node belongs according to the search service performance of each node in the search engine cluster; a shard storage submodule, which is used to, for each performance level, shard and store the commodity index of the quality level matching the performance level to the nodes belonging to the performance level, so that the nodes with higher performance levels store the shards of the commodity index with higher quality levels.

[0135] In a further embodiment, after the request response module 1300, it includes: an interest determination submodule, which is used to obtain the search products in the product search list that the user has shown interest in as interest search products; an interest description submodule, which is used to extract the user's interest description text based on the product information of each interest search product; a search sending submodule, which is used to update the search keywords according to the interest description text, obtain the target search keywords and send them to the user; a list sending submodule, which is used to trigger the construction of a corresponding product search request in response to the user confirming the target search keywords, and construct a product search list that responds to the product search request and sends it to the user.

[0136] In a further embodiment, the interest description submodule includes: a set extraction submodule, which is used to search for product information of products according to each of the interests, and extract a product common information set and a product characteristic information set; a text acquisition submodule, which is used to use a large language model to describe the user's interests according to the product common information set and the product characteristic information set, so as to obtain a corresponding generated interest description text.

[0137] In a further embodiment, the score determination submodule includes: a text experience unit, which is used to use a preset dual-tower model to identify the completeness, accuracy and compliance of the text information in the product information by the text identification branch therein, and obtain a corresponding text experience score; a picture experience unit, which is used to use the picture identification branch in the dual-tower model to identify the clarity and compliance of the picture information in the product information, and obtain a corresponding picture experience score; a basic experience unit, which is used to linearly fuse the text experience score and the picture experience score by the linear layer in the dual-tower model, and obtain a corresponding basic experience score; a forward optimization unit, which is used to use a preset product evaluation model to evaluate the richness of the product information and the selling point expression, and obtain a corresponding forward optimization score; a quality scoring unit, which is used to weightedly fuse the basic experience score and the forward optimization score to obtain a corresponding product quality score.

[0138] In order to solve the above technical problems, the present application also provides a computer device. Figure 4As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a commodity search method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the commodity search method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0139] In this embodiment, the processor is used to execute Figure 3 The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the commodity search device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0140] 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 commodity search method of any embodiment of the present application.

[0141] 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).

[0142] In summary, this application can efficiently and accurately search for products that users are interested in.

[0143] 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.

[0144] The above is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A commodity search method, characterized in that: The steps include: Obtain the product search request triggered by the user and send it to the search engine, which will parse the product search request and determine the corresponding search keywords and target product search quantity; The search engine calls the product indexes of corresponding quality levels one by one according to the search keywords and the target product search quantity and in descending order of the quality level of the product information, until a product search result related to the search keywords and meeting the target product search quantity is found from the called product indexes, and the calling is terminated and the product search result is returned; Determine the interest association score between each search product in the product search results and the user, sort the products according to the interest association score, construct a corresponding product search list and send it to the user.

2. The commodity search method according to claim 1, characterized in that: The search engine calls the product indexes of corresponding quality levels one by one according to the search keywords and the target product search quantity, and in descending order of the quality level of the product information, until a product search result related to the search keywords and meeting the target product search quantity is found from the called product indexes, and before the call is terminated and the product search result is returned, the following steps are included: Get the latest product information of each product on the e-commerce platform; For each product information, a corresponding basic experience score is determined according to the degree to which the product information satisfies the user's basic experience, and a corresponding positive optimization score is determined according to the degree to which the product information improves the user's basic experience. The product quality score of the product information is determined based on the basic experience score and the positive optimization score; Obtain the grade rating intervals associated with each quality grade divided from high to low, and determine the quality grade to which each product information belongs according to the grade rating interval to which the product quality score of each product information belongs; For each quality level, a commodity index of the quality level is constructed according to all commodity information belonging to the quality level.

3. The commodity search method according to claim 1, characterized in that: Before determining the interest association score between each search product in the product search results and the user, the following steps are included: Obtain training samples and supervisory labels in a prepared training set, wherein the training samples include key product information of a product and key user information of a user, and the supervisory labels represent whether the user has shown interest in the product; Inputting the training samples into an interest evaluation model to extract corresponding deep semantic features and obtain an interest feature vector; Classify and map the interest feature vectors to predict corresponding interest association scores; The loss value corresponding to the predicted interest association score is calculated according to the supervisory label, and the interest assessment model is updated according to the loss value until the model converges, so as to determine the interest association score between each search product in the product search results and the user.

4. The commodity search method according to claim 2, characterized in that: For each quality level, after constructing a product index of the quality level according to all product information belonging to the quality level, the following steps are included: According to the search service performance of each node in the search engine cluster, various performance levels are determined from high to low, as well as the performance level to which each node belongs; For each performance level, the commodity index of the quality level matching the performance level is stored in fragments in the nodes belonging to the performance level, so that nodes with higher performance levels store fragments of commodity indexes with higher quality levels.

5. The commodity search method according to claim 1, characterized in that: After constructing the corresponding product search list and sending it to the user, the following steps are included: Obtaining, from the commodity search list, the searched commodity for which the user has shown interest as the interest searched commodity; Extracting the user's interest description text according to the product information of each interest search product; Update the search keyword according to the interest description text, obtain the target search keyword and send it to the user; In response to the user confirming the target search keyword, a corresponding product search request is triggered, and a product search list that responds to the product search request is constructed and sent to the user.

6. The commodity search method according to claim 5, characterized in that: Extracting the user's interest description text according to the product information of each interest search product includes the following steps: Searching for product information of products according to the respective interests, and extracting a product commonality information set and a product characteristic information set; A large language model is used to describe the user's interests according to the commodity commonality information set and the commodity characteristic information set to obtain a corresponding interest description text.

7. The commodity search method according to any one of claims 1 to 6, characterized in that: For each product information, a corresponding basic experience score is determined according to the degree to which the product information satisfies the basic experience of the user, and a corresponding positive optimization score is determined according to the degree to which the product information improves the basic experience of the user. The product quality score of the product information is determined based on the basic experience score and the positive optimization score, including the following steps: A preset dual-tower model is used, in which the text identification branch identifies the completeness, accuracy and compliance of the text information in the product information to obtain a corresponding text experience score; The picture identification branch in the dual-tower model identifies the clarity and compliance of the picture information in the product information to obtain a corresponding picture experience score; The linear layer in the dual-tower model linearly fuses the text experience score and the picture experience score to obtain a corresponding basic experience score; Use a preset product evaluation model to evaluate the richness of the product information and the performance of the selling points to obtain a corresponding positive optimization score; The basic experience score and the positive optimization score are weighted and integrated to obtain a corresponding product quality score.

8. A commodity search device, characterized in that: include: The request processing module is used to obtain the product search request triggered by the user and send it to the search engine, which analyzes the product search request and determines the corresponding search keywords and the number of target product searches; A result returning module is used for the search engine to call the product indexes of corresponding quality levels one by one according to the quality level of the product information from high to low according to the search keyword and the target product search quantity, until a product search result related to the search keyword and meeting the target product search quantity is found from the called product index, and the calling is terminated and the product search result is returned; The request response module is used to determine the interest association score between each search product in the product search results and the user, sort the products according to the interest association score, and construct a corresponding product search list to send to the user.

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

10. A computer-readable storage medium, characterized in that: It stores a computer program implemented according to the method described in any one of claims 1 to 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