Commodity recall combination method and device, equipment and medium

Abstract

The invention relates to a commodity recall combination method and device, equipment and a medium in the technical field of e-commerce, and the method comprises the steps: determining a plurality of recall sources associated with a recommendation scene and recommendation effect scores of the recall sources for various recommendation scenes corresponding to an independent station shop, adding the recall source with the highest score to a target recall combination, adding the remaining recall sources to a candidate recall set, starting iteration, combining each recall source in the set with a target recall combination, determining the recommendation value improvement degree of each correspondingly obtained candidate recall combination compared with the target recall combination, and if the highest recommendation value improvement degree reaches the standard, returning to the candidate recall set; rejecting a combined recall source in a candidate recall combination corresponding to the improvement degree from the set, updating the candidate recall combination into a target recall combination, and continuing iteration; if the highest recommendation value improvement degree does not reach the standard, iteration is ended, and a current target recall combination is obtained. According to the method, the recall combination with the highest recommendation value can be fully mined for the scene.

Description

Technical Field

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

[0002] An independent website usually has its own domain name and an online store built on this basis. The store has a high degree of autonomy and independence in operation. Each store has its own type, and there may be significant differences in operating cycle, business category, target market, traffic channels, etc.

[0003] In traditional technology, various recall sources are usually added to serve the various product recommendation needs of each independent station store. However, the creation cost and online deployment of recall sources require a lot of manpower and material resources, with low economic benefits and high consumption of computing resources.

[0004] In view of the shortcomings of traditional practices, the applicant has been engaged in research in related fields for a long time and has taken a different approach to solve industry problems in the field of e-commerce 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 recall combination method and its corresponding device, computer equipment, and computer program product.

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

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

[0008] For each preset recommendation scenario corresponding to the independent station store, determine the multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores, add the recall source with the highest recommendation effectiveness score to the target recall combination, and add the remaining recall sources to the candidate recall set;

[0009] Initiate an iterative process, combine each recall source in the candidate recall set with the target recall combination, obtain multiple corresponding candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared to the target recall combination;

[0010] When the highest recommendation value improvement is greater than or equal to the preset threshold, the recall source combined in the candidate recall combination corresponding to the recommendation value improvement is removed from the candidate recall set, and the candidate recall combination is updated to the target recall combination, and the iterative process continues;

[0011] When the highest recommendation value improvement is less than the preset threshold, the iteration process ends and the current target recall combination is obtained to prepare for product recommendations serving the recommendation scenario.

[0012] On the other hand, a product recall combination device provided to meet one of the purposes of the present application includes a data construction module, an iteration start module, an iteration continuation module and an iteration end module, wherein the data construction module is used to determine multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores for various preset recommendation scenarios corresponding to the independent station store, add the recall source with the highest recommendation effectiveness score to the target recall combination, and add the remaining recall sources to the candidate recall set; the iteration start module is used to start the iteration process, combine each recall source in the candidate recall set with the target recall combination respectively, obtain the corresponding multiple candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared with the target recall combination; the iteration continuation module is used to, when the highest recommendation value improvement is greater than or equal to a preset threshold, remove the combined recall sources in the candidate recall combination corresponding to the recommendation value improvement from the candidate recall set, and update the candidate recall combination to the target recall combination to continue the iteration process; the iteration end module is used to, when the highest recommendation value improvement is less than the preset threshold, end the iteration process, obtain the current target recall combination, and prepare to serve the product recommendation of the recommendation scenario.

[0013] On the other hand, a computer device is provided to meet one of the purposes of the present application, including 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 recall combination method described in the present application.

[0014] On the other hand, a computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0015] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0016] For each recommendation scenario of the independent station store service, this application selects the recall source with the best recommendation performance in the current scenario based on multiple recall sources related to the recommendation scenario and their recommendation effectiveness scores, and directly adds it to the target recall combination to create a starting point, ensuring that the most relevant and valuable product recommendations can be provided on this basis. Secondly, by adding the remaining recall sources to the candidate recall set, other potentially effective recall sources are retained to prepare for subsequent iterative exploration for optimization.

[0017] In the iterative process, the recall sources in the candidate recall set are combined with the target recall combination, and the recommendation value improvement of these combinations is evaluated. This can quantify the impact of different recall source combinations on the recommendation effect, and thus make a corresponding decision on whether to adopt them if the improvement is sufficient. When it is found that the recommendation value improvement of a candidate recall combination meets the standard, the combined recall sources in the combination are included in the target recall combination, which indicates that the recall source has a significant positive optimization impact on the recommendation effect and is worthy of being included.

[0018] Finally, when it is found that the combination with each recall source cannot significantly improve the recommendation value, the iterative process ends. At this time, the target recall combination is confirmed to have been continuously optimized until it cannot be optimized any further. It can reasonably utilize the existing optimal part of the recommendation recall resources, and thus is expected to provide the best recommendation effect for the recommendation scenario. 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 product recall combination method of the present application;

[0022] Figure 3 This is a principle block diagram of the product recall assembly 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 online services such as cloud computing services, 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 data processing system. Therefore, the data processing system can run on any one or more servers of the e-commerce platform's cluster, so as to realize various functions by using the various commodity data provided by the e-commerce platform.

[0039] A product recall combination 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 product recall combination method of the present application, in its typical embodiment, comprises the following steps:

[0041] Step S1100: for various preset recommendation scenarios corresponding to the independent website store, multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores are determined, the recall source with the highest recommendation effectiveness score is added to the target recall combination, and the remaining recall sources are added to the candidate recall set;

[0042] In independent station stores, at least one recall source is usually pre-configured for various recommendation scenarios of its services. Secondly, the same recall source can be applicable to multiple recommendation scenarios. The recall source is applicable to the corresponding recommendation scenario and is implemented based on a recall strategy and / or recall channel. For example, for recommendations whose recommendation scenario is the user's click behavior, since the user clicks on the product because of interest, in order to achieve recommendations that meet the user's expectations, the corresponding configuration of various recall sources may be similar product title recall, similar product category recall, similar product image recall, similar product information recall, etc., to recall products similar to the product clicked by the user; for recommendations whose recommendation scenario is the homepage, in order to show users the products that they may be interested in, the corresponding configuration of various recall sources may be user behavior recall, collaborative filtering recall, store high click recall, store high add-on recall, store high order recall, user region recall, etc., to recall products that users are interested in.

[0043] The recommendation effectiveness score quantitatively represents the effectiveness of the recall source in recommending the product to the user after recalling it in the recommendation scenario, and can be any one or a weighted fusion value of any multiple of the corresponding click-through rate, conversion rate, click-through conversion rate, AUC, recall rate, precision recall rate, etc. Those skilled in the art can flexibly determine the recommendation effectiveness score based on the disclosure herein, or can implement it according to the disclosure of some subsequent embodiments.

[0044] After the recall source recalls the product in the recommendation scenario, it recommends it to the user. Specifically, the recall source recalls all the products from the independent station store, usually quantitatively represents the recall score corresponding to each product, and then selects multiple products with relatively high recall scores as recommended products, which are pushed to the user. The recall score represents the value of the product being recalled by the corresponding recall source, and the value refers to the degree of match with the recall target of the recall source. It is not difficult to understand that different recall sources have their own recall targets and their respective recall targets are different. For example, the recall target of the recall of similar product titles is to recall products that are similar to the product titles of the specified product. In this regard, the value of the product depends on the degree of similarity between the product title of the product and the product title of the specified product. The quantitatively represented degree of similarity is the recall score. The number of preferred products can be set as needed by technicians in this field.

[0045] Step S1200, start the iterative process, combine each recall source in the candidate recall set with the target recall combination, obtain a corresponding plurality of candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared with the target recall combination;

[0046] From then on, the iteration begins, traversing each recall source in the candidate recall set, combining each recall source with the target recall combination to obtain multiple corresponding candidate recall combinations, exhaustively obtaining various combinations to ensure the comprehensiveness of the combination exploration and mining, and then further examining the recommendation value improvement of each candidate recall combination compared to the target recall combination before the combination. The recommendation value improvement quantifies the improvement in the recommendation value of the candidate recall combination compared to the target recall combination in the recommendation scenario. The recommendation value measurement index can be any one or more of the effectiveness of the recommended product, the degree of diversity of the recommended product, the computational overhead of the recommended product, etc., and the corresponding improvement in the recommendation value can be reflected in any one or more of the improved effectiveness, the higher degree of diversity, the lower computational overhead, etc., and the corresponding improvement can be quantified as a ratio or difference. Those skilled in the art can flexibly and flexibly determine the recommendation value improvement based on the disclosure herein, and can also implement it according to the disclosure of some subsequent embodiments.

[0047] Regarding the recall combination, in the recommendation scenario, the recalled products are recommended to the user. Specifically, in one embodiment, each recall source in the recall combination first recalls all products in the independent station store, determines the recall score obtained by each recall source for each product in the store, and then for each product, multiplies each recall score of the product by its weight and then adds them up to get the recommendation score. Further, based on the recommendation score sorting, multiple products with higher recommendation scores are screened out as recommended products and sent to the user. The weights can be pre-set by those skilled in the art as needed. The number of screened products can be set by those skilled in the art as needed.

[0048] In another embodiment, each recall source in the recall combination recalls all products in the independent station store, quantifies the corresponding recall score of each product, and then selects multiple products with relatively high recall scores as candidate products, and collects all candidate products to form a candidate product set of the recall source. Furthermore, multiple candidate products with high corresponding recall scores are selected from each candidate product set as recommended products, and different recommended products are sent to users. That is to say, if there are duplicate products among these recommended products, a corresponding non-duplicate recommended product can be sent. The number of products to be screened can be set as needed by those skilled in the art.

[0049] Step S1300: When the highest recommendation value improvement is greater than or equal to a preset threshold, the recall source combined in the candidate recall combination corresponding to the recommendation value improvement is removed from the candidate recall set, and the candidate recall combination is updated to the target recall combination, and the iterative process continues;

[0050] At this time, it means that the current highest recommendation value improvement is high enough, and the candidate recall combination corresponding to the recommendation value improvement has a further considerable improvement in recommendation value compared to before the combination. Therefore, the combined recall source in the candidate recall combination is removed from the candidate recall set, and the candidate recall combination is updated to the target recall combination, so that the possibility of further exploring other combinations based on the target recall combination is further explored in the future. The preset threshold can be set by those skilled in the art as needed.

[0051] Step S1400: When the highest recommendation value improvement is less than a preset threshold, the iterative process ends and the current target recall combination is obtained to prepare for product recommendations serving the recommendation scenario.

[0052] This means that the current highest recommendation value improvement is not high enough, and the candidate recall combination corresponding to the recommendation value improvement has not been significantly improved compared to the previous combination. It also proves that the current target recall combination is obtained through continuous exploration and iteration to continuously improve the recommendation value until the improvement is not enough. The target recall combination is obtained, and the iteration is ended. The target recall combination is used to recommend products in the recommendation scenario.

[0053] 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:

[0054] For each recommendation scenario of the independent station store service, this application selects the recall source with the best recommendation performance in the current scenario based on multiple recall sources related to the recommendation scenario and their recommendation effectiveness scores, and directly adds it to the target recall combination to create a starting point, ensuring that the most relevant and valuable product recommendations can be provided on this basis. Secondly, by adding the remaining recall sources to the candidate recall set, other potentially effective recall sources are retained to prepare for subsequent iterative exploration for optimization.

[0055] In the iterative process, the recall sources in the candidate recall set are combined with the target recall combination, and the recommendation value improvement of these combinations is evaluated. This can quantify the impact of different recall source combinations on the recommendation effect, and thus make a corresponding decision on whether to adopt them if the improvement is sufficient. When it is found that the recommendation value improvement of a candidate recall combination meets the standard, the combined recall sources in the combination are included in the target recall combination, which indicates that the recall source has a significant positive optimization impact on the recommendation effect and is worthy of being included.

[0056] Finally, when it is found that the combination with each recall source cannot significantly improve the recommendation value, the iterative process ends. At this time, the target recall combination is confirmed to have been continuously optimized until it cannot be optimized any further. It can reasonably utilize the existing optimal part of the recommendation recall resources, and thus is expected to provide the best recommendation effect for the recommendation scenario.

[0057] In a further embodiment, step S1200, determining the recommendation value improvement of each candidate recall combination compared to the target recall combination, includes the following steps:

[0058] Step S1210: for each candidate recall combination, determine the first recommended product set corresponding to the target recall combination and its recommendation effectiveness score and recommendation computing cost index, as well as the second recommended product set corresponding to the candidate recall combination and its recommendation effectiveness score and recommendation computing cost index;

[0059] In one embodiment, the first recommended product set responds to a product recommendation event triggered by a user, uses each recall source in the target recall combination to recall all products in the independent station store, quantifies the recall score corresponding to each product, and then selects multiple products with relatively high recall scores as candidate products, and collects all candidate products to form a candidate product set of the recall source. Furthermore, multiple candidate products with relatively high corresponding recall scores are selected from each candidate product set as recommended products, and different recommended products are collected to form a first recommended product set and sent to the user. That is to say, if there are duplicate products among the recommended products, a corresponding non-duplicate recommended product can be sent. The number of products to be screened can be set as needed by those skilled in the art.

[0060] The second recommended product set responds to the product recommendation event triggered by the user, uses each recall source in the candidate recall combination to recall all products in the independent station store, quantifies the recall score corresponding to each product, and then selects multiple products with relatively high recall scores as candidate products, and collects all candidate products to form the candidate product set of the recall source. Furthermore, multiple candidate products with high corresponding recall scores are selected from each candidate product set as recommended products, and different recommended products are collected to form two recommended product sets and sent to the user. That is to say, if there are duplicate products among the recommended products, a corresponding non-duplicate recommended product can be sent. The number of products to be screened can be set as needed by those skilled in the art.

[0061] The recommendation effectiveness score quantitatively represents the effectiveness of the corresponding recommended product set in recommending to the user in the recommendation scenario, and can be any one or a weighted fusion value of any multiple of the corresponding click-through rate, conversion rate, click-through conversion rate, AUC, recall rate, precision rate, etc. Those skilled in the art can flexibly determine the recommendation effectiveness score based on the disclosure herein.

[0062] The recommendation computing overhead indicator includes computing resources and / or computing time required to generate the corresponding recommended product set. The computing resources may be any one or more of memory usage, CPU usage, GPU usage, memory usage, etc.

[0063] Step S1220: determining the recommendation effectiveness improvement degree according to the recommendation effectiveness score of the first recommended product set and the recommendation effectiveness score of the second recommended product set;

[0064] A ratio obtained by subtracting the recommendation effectiveness score of the first recommended product set from the recommendation effectiveness score of the second recommended product set is divided by the recommendation effectiveness score of the first recommended product set to obtain a recommendation effectiveness improvement degree.

[0065] Step S1230: determining a difference lift according to the total number of different commodities between the first recommended commodity set and the second recommended commodity set and the total number of commodities in the first recommended commodity set;

[0066] The different products between the first recommended product set and the second recommended product set are determined as differential products, the number of these differential products is accumulated to obtain the total number of differential products, and the number of all products in the first recommended product set is accumulated to obtain the total number of products. Then, the ratio of the total number of differences divided by the total number of products is used as the difference improvement degree.

[0067] Step S1240: determining a computing cost increase degree according to the recommended computing cost index of the first recommended product set and the recommended computing cost index of the second recommended product set;

[0068] Calculate the value of each dimension in the recommended computing cost index of the second recommended product set, subtract the value of each same dimension in the recommended computing cost index of the first recommended product set, and get the difference of each dimension. Then calculate the difference of each dimension and divide it by the value of each same dimension in the recommended computing cost index of the first recommended product set to get the ratio of each dimension. The sum of these ratios is used as the computing cost increase.

[0069] Step S1250: Determine the recommendation value improvement degree according to the recommendation effectiveness improvement degree, the difference improvement degree and the calculation cost increase degree.

[0070] In one embodiment, the recommendation effectiveness improvement, difference improvement, and computational cost increase are multiplied by their respective weights and then added together to obtain the recommendation value improvement. The weights can be set by those skilled in the art as needed.

[0071] In another embodiment, the recommendation effectiveness improvement, difference improvement, and computational cost increase are added together and then divided by three to obtain the recommendation value improvement.

[0072] In this embodiment, by examining each candidate recall combination compared to the target recall combination, the three aspects of the effectiveness of recommending products, the degree of diversity of recommended products, and the computational overhead of recommended products are comprehensively considered to determine the improvement in recommendation value between the candidate recall combination and the target recall combination. This ensures that the smaller the computational overhead of the recommended products, the richer they are, and the more effective they are, the higher the corresponding recommendation value.

[0073] In a further embodiment, after step S1400, obtaining the current target recall combination, the following steps are included:

[0074] Step S1510: respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, obtain the candidate product set respectively recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product therein;

[0075] On the one hand, any of the various actions taken by users in the independent station store, such as browsing, searching, clicking, adding to the shopping cart, purchasing, visiting, etc., can trigger product recommendation events. On the other hand, any of the specific pages visited by users in the independent station store, such as the home page, order settlement page, order placement page, shopping cart page, search page, browsing history page, etc., can trigger product recommendation events.

[0076] In an independent website store, for various different product recommendation events that occur therein, recommendation scenarios are pre-configured. The recommendation scenarios may refer to the recommendation of pages and / or behaviors that trigger the corresponding product recommendation events, or may refer to the recommendation of activities formulated based on the operation and / or marketing conditions of the independent website store. In addition, an event identifier is pre-set for each product recommendation event. The event identifier is used to uniquely refer to the corresponding product recommendation event to distinguish it from other product recommendation events, and can be flexibly implemented by those skilled in the art.

[0077] In independent station stores, at least one recall source is usually pre-configured for various recommendation scenarios of its services. Secondly, the same recall source can be applicable to multiple recommendation scenarios. The recall source is applicable to the corresponding recommendation scenario and is implemented based on a recall strategy and / or recall channel. For example, for recommendations whose recommendation scenario is the user's click behavior, since the user clicks on the product because of interest, in order to achieve recommendations that meet the user's expectations, the corresponding configuration of various recall sources may be similar product title recall, similar product category recall, similar product image recall, similar product information recall, etc., to recall products similar to the product clicked by the user; for recommendations whose recommendation scenario is the homepage, in order to show users the products that they may be interested in, the corresponding configuration of various recall sources may be user behavior recall, collaborative filtering recall, store high click recall, store high add-on recall, store high order recall, user region recall, etc., to recall products that users are interested in.

[0078] When a user in an independent station store triggers a product recommendation event, the user's terminal device carrying the independent station store constructs a product recommendation request carrying the event identifier of the product recommendation event and sends it to the server.

[0079] The server receives the product recommendation request, responds to the event, takes the user as the target user, and determines the recommendation scenario pre-configured for the event according to the unique identifier as the target recommendation scenario.

[0080] Step S1520: Initialize and construct effectiveness distribution corresponding to each candidate product set according to the recommendation effectiveness score of each recall source;

[0081] The total number of historical exposures corresponding to the candidate product set recalled by each recall source is obtained, and the total number of historical exposures is the sum of the total number of historical exposures of each candidate product in the corresponding candidate product set. Based on the algorithm principle of Thompson sampling, for each recall source, the total number of historical exposures corresponding to the candidate product set recalled by the recall source is multiplied by the recommendation effectiveness score of the recall source to obtain the effectiveness prior value, which is used as α in Beta (α, β). The total number of historical exposures is subtracted from the effectiveness prior value to obtain the effectiveness prior value, which is used as β in Beta (α, β). The Beta (α, β) is constructed with the α and β as the effectiveness distribution.

[0082] Step S1530: updating the respective achievement distributions accordingly according to the historical user behaviors of the target user and the respective users in the user group to which the target user belongs and the respective candidate product sets;

[0083] In one embodiment, a clustering algorithm is used to cluster all users in the independent station store based on user feature information of all users, so as to determine at least one user group and the user group to which each user belongs. The clustering algorithm can be any one of the K-Means algorithm, DBSCAN algorithm, etc., and technical personnel in this field can flexibly implement it.

[0084] The historical user behaviors of the target user and each user in the user group to which he belongs in the independent station store and corresponding to the candidate product set refer to whether the user makes a conversion behavior on the candidate product in the candidate product set after the candidate product in the candidate product set is exposed to the user. Thus, all the historical user behaviors of the target user and each user in the user group to which he belongs are obtained. For each historical user behavior, if the historical user behavior is a conversion behavior, add one to α in the effectiveness distribution; if the historical user behavior is not a conversion behavior, add one to β in the effectiveness distribution. In this way, real-time updating of the effectiveness distribution is achieved.

[0085] Step S1540: randomly sampling the corresponding expected effectiveness probabilities for the respective effectiveness distributions, and adjusting the recall scores of the candidate products in each candidate product set according to the respective expected effectiveness probabilities to obtain a recommendation score;

[0086] For each achievement distribution, a random number is randomly generated as the expected probability of achievement based on its deformation parameters α and β using the achievement distribution.

[0087] For each candidate product set, the recall score of each candidate product in the set is multiplied by the expected probability of effectiveness corresponding to the set to obtain the recommendation score of each candidate product.

[0088] Step S1550: Filter out a plurality of recommended products based on the recommendation score sorting and push them to the target user.

[0089] All the recalled candidate products are sorted in descending order of recommendation scores, and multiple different candidate products with the highest rankings are selected as recommended products. Then, a recommendation display page of the target recommendation scenario is obtained, and the sorted recommended products are placed in the page in order, that is, the higher the recommended products are ranked, the closer they are to the top of the page, so that users can browse products with better recommended effects first based on their top-down browsing habits. The recommendation display page is pushed to the target user, and the page is loaded and displayed, so that the recommended products are exposed to the target user, and the products are recommended to the target user.

[0090] In this embodiment, the effectiveness distribution of the current recalled products is constructed based on the prior recommendation effectiveness experience, and then the effectiveness distribution is updated based on the a posteriori recommendation effectiveness data. The corresponding expected effectiveness probability is sampled based on this to adjust the recall score of the current recalled product to obtain the recommendation score, and the recommended products with higher scores are selected and sent to the user, so that the recommendation results are more in line with the user's real-time preferences and behavioral trends, which not only improves the accuracy of the recommendation, but also enhances the user experience, making it more likely that users will see the products they are interested in.

[0091] In a further embodiment, step S1100, determining multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores, includes the following steps:

[0092] Step S1110: based on each recall source associated with the recommendation scenario, recall corresponding target products for the user's product recommendation event, and push the target products recalled by each recall source to the user;

[0093] In an independent station store, at least one recall source is usually pre-configured for various recommendation scenarios, thereby obtaining at least one recall source configured for each recommendation scenario. For each recommendation scenario, in response to the product recommendation event of the user in the recommendation scenario, each recall source in the recommendation scenario is called for each product recommendation event. For each recall source, the recall source determines the recall score of each product from all the products in the independent station store, sorts them in descending order according to the recall score, selects the top ranked products as target products, and then recalls these target products. After that, these recalled target products are sent to the user who triggered the product recommendation event.

[0094] Step S1120: Determine the recommendation effectiveness score of each recall source based on whether the user makes a conversion behavior on the target product after the target product is exposed to the corresponding user.

[0095] After each target product is exposed to the corresponding user, monitor whether the user has made a conversion behavior on the target product. When the user makes a conversion behavior on the target product, the target product's performance indicator is generated as the performance reaches the standard; when the user does not make a conversion behavior on the target product, the target product's performance indicator is generated as the performance does not reach the standard. The conversion behavior can be any one or more of click, add to cart, purchase, repeat purchase, etc.

[0096] For each recommendation scenario, based on the effectiveness indicators of the target products recalled by each recall source in the recommendation scenario, the ratio of the total number of target products that meet the effectiveness standards divided by the total number of target products that do not meet the effectiveness standards among all target products recalled by each recall source is determined as the recommendation effectiveness score of the recall source in the recommendation scenario.

[0097] In this embodiment, each recall source applicable to the recommendation scenario is deployed to recall the corresponding target product and send it to the user, and then the recommendation effectiveness score of the recall source is determined based on whether the user makes a conversion behavior. This can accurately quantify the effectiveness of the recall source in recommending the product to the user after recalling it in the corresponding recommendation scenario.

[0098] In a further embodiment, after step S1400, obtaining the current target recall combination, the following steps are included:

[0099] Step S2500: responding to the recommendation computing overhead overload event of the recommendation scenario, obtaining the target recall combination corresponding to each independent station store serving the recommendation scenario;

[0100] It can be understood that for various recommendation scenarios, the target recall combination adopted by each independent station store in the recommendation scenario can be determined, and the recommendation computing overhead index required for product recommendation based on the target recall combination can be determined. Therefore, for e-commerce platforms, they usually provide computing resources for all online services of each independent station store, so the recommendation computing overhead indicators required for these independent station stores to use the corresponding target recall combination to implement product recommendations in various recommendation scenarios can be aggregated. When the sum of the recommendation computing overhead indicators corresponding to the same recommendation scenario exceeds the preset threshold, the recommendation computing overhead of the recommendation scenario is overloaded. In response to the corresponding event, the target recall combination adopted by each independent station store in the recommendation scenario is obtained. The preset threshold can be set as needed by a technician in this field according to the disclosure herein.

[0101] Step S2510, determining whether each target recall combination meets a preset similarity condition, and grouping the target recall combinations that meet the preset similarity condition into the same recall combination group;

[0102] Determine the total number of recall sources of the differences between each target recall combination. When the total number of recall sources of the differences is less than or equal to the preset number, it is confirmed that the corresponding two target recall combinations belong to the same category; when the total number of recall sources of the differences exceeds the preset number, it is confirmed that the corresponding two target recall combinations do not belong to the same category. Thus, target recall combinations belonging to the same category are grouped into the same group, and each group is obtained as a recall combination group. The preset number can be set by a person skilled in the art as needed.

[0103] Step S2520: for each recall combination group, select multiple target recall combinations with relatively low recommendation value enhancement and the highest target recall combination from the recall combination group, and update the highest target recall combination to the target recall combination corresponding to the recommendation scenario served by the independent station store corresponding to each of the lower target recall combinations.

[0104] Taking a single recall combination group as an example, all target recall combinations are sorted from low to high according to the recommendation value enhancement corresponding to each target recall combination in the recall combination group, and multiple target recall combinations with higher rankings and the target recall combination with the last ranking are screened out, and the target recall combination with the last ranking is used to serve the target recall combination corresponding to the independent station store corresponding to each target recall combination with the higher rankings to serve the recommendation scenario, so that the total number of different target recall combinations deployed by the e-commerce platform for all independent station stores serving the recommendation scenario is reduced, so as to reduce the recommendation operation load from overload to normal. Those skilled in the art can set the number of target recall combinations with relatively low recommendation enhancement value to be screened as needed according to the disclosure herein.

[0105] In this embodiment, when the recommendation computing overhead of a recommendation scenario is overloaded, the total number of different target recall combinations that the e-commerce platform needs to deploy for the recommendation scenario is appropriately adjusted, so that the overall recommendation computing overhead can be reduced while minimizing the impact on the overall recommendation effectiveness.

[0106] In a further embodiment, after step S1400, obtaining the current target recall combination, the following steps are included:

[0107] Step S1501: respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, obtain the candidate product set respectively recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product therein;

[0108] When a user in an independent station store triggers a product recommendation event, the user's terminal device carrying the independent station store constructs a product recommendation request carrying the event identifier of the product recommendation event and sends it to the server.

[0109] The server receives the product recommendation request, responds to the event, takes the user as the target user, and determines the recommendation scenario pre-configured for the event according to the unique identifier as the target recommendation scenario.

[0110] Step S1502: Determine the conversion value score of each recalled candidate product associated with the target user;

[0111] The user characteristic information of the target user and the product characteristic information of each candidate product are obtained, and a conversion value score of each candidate product is determined correspondingly using a conversion evaluation model that has been pre-trained to convergence based on the user characteristic information and each product characteristic information as input. The conversion evaluation model is trained to acquire the ability to reliably and accurately predict the conversion value score including any one or more of the click-through rate, conversion rate, and click-to-conversion rate based on the input user characteristic information and product characteristic information.

[0112] Step S1503: determining a recommendation score according to the recall score and the conversion value score of each candidate product, and sorting and selecting a plurality of recommended products based on the recommendation score to push to the target user.

[0113] Furthermore, for each candidate product, the recall score and conversion value score of the candidate product are multiplied by their respective weights and then added to obtain a recommendation score. All recalled candidate products are sorted in descending order of recommendation scores, and multiple candidate products with the highest rankings are selected as recommended products. Then, the recommendation display page of the target recommendation scenario is obtained, and each recommended product is placed on the recommendation display page in sequence and sent to the target user. The number of candidate products to be screened, the weight of the recall score, and the weight of the conversion value score can all be pre-set by technicians in this field according to business needs.

[0114] In this embodiment, when recommending products, the target users and target recommendation scenarios for the recommendations are clearly defined, and each recall source in the target recall combination corresponding to the scenario is called to recall the candidate product set respectively, and the recall score of each recalled candidate product and the recommendation score based on the conversion value score of the associated target user are determined, and the recommended products with higher recommendation scores are selected and sent to the users. It can be seen that for the specified recommendation scenario and user, the value of product recommendation is quantified based on recall improvement and associated users, so as to achieve personalized recommendations based on accurately assessed value to enhance the user shopping experience and help improve the conversion rate.

[0115] In a further embodiment, before step S1501, responding to the product recommendation event of the independent station store, the following steps are included:

[0116] Step S3500: obtaining training samples and supervisory labels in a prepared training set, wherein the training samples include product feature information of a product and user feature information of a user, and the supervisory labels represent whether the user has made a conversion behavior with respect to the product;

[0117] The product feature information includes all pre-processed texts describing the product 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 any one or more of the number of historical exposures, clicks, purchases, etc. of the product.

[0118] The user characteristic information includes the pre-processed user portrait, and / or, when the user makes a conversion behavior on the product, the description text of the product 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 products, adding to shopping carts, purchasing products, clicking products, collecting products, and commenting on products in independent station stores.

[0119] 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.

[0120] As for the training samples and their supervisory labels prepared in advance in the training set, specifically, by monitoring the interaction between users and commodities in the independent station store, when the user makes a conversion behavior on the commodity, the user feature information of the user and the feature information of the commodity are collected as training samples, and the supervisory label of the training sample is marked as representing that the user makes a conversion behavior on the commodity, 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 make a conversion behavior on the commodity, the user feature information of the user and the feature information of the commodity are collected as training samples, and the supervisory label of the training sample is marked as representing that the user does not make a conversion behavior on the commodity, 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.

[0121] Step S3510: Input the training sample into the conversion evaluation model to extract corresponding deep semantic features and obtain a conversion feature vector;

[0122] It is recommended to use ESMM in the selection of the conversion evaluation model. Other variant models such as LR, DeepFM model, etc. that can achieve the conversion value scoring function between users and their associated products in the present application can also be replaced by it.

[0123] In one embodiment, a conversion evaluation model based on ESSM is adopted, and each supervision label in the training set includes the click rate, click-through rate, and conversion rate corresponding to the corresponding training sample. The training samples are input into the conversion evaluation model, and the shared embedding layer in the conversion 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 conversion 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 conversion feature vector for predicting the click-through rate; the second branch is that the pooling layer in the conversion prediction tower in the conversion 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 conversion feature vector for predicting the conversion rate.

[0124] Step S3520: classify and map the conversion feature vector to predict the corresponding conversion value score;

[0125] Following the first branch, the multi-layer perceptron (MLP) in the click prediction tower in the conversion evaluation model performs a nonlinear transformation on the conversion 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 conversion evaluation model performs a nonlinear transformation on the conversion 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 conversion 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 conversion value scores corresponding to the training samples.

[0126] Step S3530: Calculate the loss value corresponding to the conversion value score according to the supervision label, and update the conversion evaluation model according to the loss value until the model converges to determine the conversion value score of the recalled product associated with the user who triggered the recall.

[0127] Call the preset cross entropy loss function, calculate the cross entropy loss values ​​corresponding to the predicted click rate, predicted conversion rate, and predicted click-to-purchase rate according to the supervisory labels of the training samples, and when each cross entropy loss value reaches a preset threshold, it indicates that the conversion evaluation model has been trained to a convergence state, thereby terminating the conversion evaluation model training; when any cross entropy loss value does not reach the preset threshold, it indicates that the conversion evaluation model has not converged, and then implement a gradient update on the model 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 implement iterative training on the conversion 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.

[0128] In this embodiment, the conversion evaluation model is trained until it reaches a convergence state, so that it can learn the ability to reliably and accurately predict the conversion value score including any one or more of the click-through rate, conversion rate, and click-to-conversion rate based on the input user feature information and product feature information.

[0129] See also Figure 3, a product recall combination device provided to meet one of the purposes of the present application is a functional embodiment of the product recall combination method of the present application. On the other hand, the device is a product recall combination device provided to meet one of the purposes of the present application, including a data construction module 1100, an iteration start module 1200, an iteration continue module 1300 and an iteration end module 1400, wherein the data construction module 1100 is used to determine multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores for various preset recommendation scenarios corresponding to the independent station store, add the recall source with the highest recommendation effectiveness score to the target recall combination, and add the remaining recall sources to the candidate recall set; the iteration start module 1200 is used Start the iterative process, combine each recall source in the candidate recall set with the target recall combination respectively, obtain multiple corresponding candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared with the target recall combination; the iteration continuation module 1300 is used to remove the recall sources combined in the candidate recall combination corresponding to the recommendation value improvement from the candidate recall set when the highest recommendation value improvement is greater than or equal to a preset threshold, and update the candidate recall combination to the target recall combination to continue the iterative process; the iteration end module 1400 is used to end the iterative process when the highest recommendation value improvement is less than the preset threshold, and obtain the current target recall combination to prepare for the product recommendation serving the recommendation scenario.

[0130] In a further embodiment, the iterative start module 1200 includes: a numerical determination submodule, which is used to determine, for each candidate recall combination, the first recommended product set corresponding to the target recall combination and its recommendation effectiveness score and recommendation computing cost index, as well as the second recommended product set corresponding to the candidate recall combination and its recommendation effectiveness score and recommendation computing cost index; a recommendation effectiveness improvement submodule, which is used to determine the recommendation effectiveness improvement degree according to the recommendation effectiveness score of the first recommended product set and the recommendation effectiveness score of the second recommended product set; a difference improvement submodule, which is used to determine the difference improvement degree according to the total number of difference products between the first recommended product set and the second recommended product set, and the total number of products in the first recommended product set; a computing cost increase submodule, which is used to determine the computing cost increase degree according to the recommendation computing cost index of the first recommended product set and the recommendation computing cost index of the second recommended product set; a recommendation value improvement submodule, which is used to determine the recommendation value improvement degree according to the recommendation effectiveness improvement degree, the difference improvement degree and the computing cost increase degree.

[0131] In a further embodiment, after the iteration end module 1400, it includes: a first event response submodule, which is used to respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, and obtain the candidate product sets recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product therein; a distribution initialization submodule, which is used to initialize and construct the effectiveness distribution corresponding to each candidate product set according to the recommendation effectiveness score of each recall source; a distribution update submodule, which is used to update each effectiveness distribution according to the historical user behavior corresponding to each candidate product set associated with the target user and each user in the user group to which he belongs; a first score determination submodule, which is used to randomly sample the corresponding expected effectiveness probability for each effectiveness distribution, and adjust the recall score of the candidate product in each candidate product set according to each expected effectiveness probability to obtain a recommendation score; a first user push submodule, which is used to select multiple recommended products based on the recommendation score sorting and push them to the target user.

[0132] In a further embodiment, the data construction module 1100 includes: a second user push sub-module, which is used to recall the corresponding target product for the user's product recommendation event based on each recall source associated with the recommendation scenario, and push the target products recalled by each recall source to the user; a second score determination sub-module, which is used to determine the recommendation effectiveness score of each recall source based on whether the user makes a conversion behavior on the target product after the target product is exposed to the corresponding user.

[0133] In a further embodiment, after the iteration end module 1400, it includes: a second event response submodule, which is used to respond to the recommendation operation overhead overload event of the recommendation scenario, and obtain the target recall combination corresponding to each independent station store serving the recommendation scenario; a combination aggregation submodule, which is used to determine whether each target recall combination satisfies the preset similar conditions, and group the target recall combinations that meet the preset similar conditions into the same recall combination group; a combination update submodule, which is used to screen out multiple target recall combinations with relatively low recommendation value improvement and the highest target recall combination from each recall combination group, and update the highest target recall combination to the target recall combination corresponding to the independent station store serving the recommendation scenario corresponding to each lower target recall combination.

[0134] In a further embodiment, after the iteration end module 1400, it includes: a third event response sub-module, which is used to respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, obtain the candidate product set recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product therein; a third score determination sub-module, which is used to determine the conversion value score of each recalled candidate product associated with the target user; a third user push sub-module, which is used to determine the recommendation score according to the recall score and conversion value score of each candidate product, and sort and filter multiple recommended products based on the recommendation score and push them to the target user.

[0135] In a further embodiment, the third event response submodule includes: a training preparation submodule for obtaining training samples and supervisory labels in a prepared training set, wherein the training samples include product feature information of the product and user feature information of the user, and the supervisory label represents whether the user has made a conversion behavior on the product; a feature representation submodule for inputting the training samples into the conversion evaluation model to extract corresponding deep semantic features and obtain a conversion feature vector; a classification mapping submodule for classifying and mapping the conversion feature vector to predict a corresponding conversion value score; a model training submodule for calculating a loss value corresponding to the conversion value score according to the supervisory label, and updating the conversion evaluation model according to the loss value until the model converges, so as to determine the conversion value score of the user associated with the recalled product that triggers the recall.

[0136] In order to solve the above technical problems, the present application also provides a computer device. Figure 4 As 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 product recall combination 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 product recall combination 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.

[0137] 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 product recall assembly device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0138] 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 product recall combination method of any embodiment of the present application.

[0139] 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).

[0140] In summary, this application can fully explore the recall combination with the highest recommendation value for each scenario, and is expected to achieve the expected recommendation results.

[0141] 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.

[0142] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A product recall combination method, characterized in that: The steps include: For each preset recommendation scenario corresponding to the independent station store, determine the multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores, add the recall source with the highest recommendation effectiveness score to the target recall combination, and add the remaining recall sources to the candidate recall set; Initiate an iterative process, combine each recall source in the candidate recall set with the target recall combination, obtain multiple corresponding candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared to the target recall combination; When the highest recommendation value improvement is greater than or equal to the preset threshold, the recall source combined in the candidate recall combination corresponding to the recommendation value improvement is removed from the candidate recall set, and the candidate recall combination is updated to the target recall combination, and the iterative process continues; When the highest recommendation value improvement is less than the preset threshold, the iteration process ends and the current target recall combination is obtained to prepare for product recommendations serving the recommendation scenario.

2. The product recall combination method according to claim 1, characterized in that: Determining the recommendation value improvement of each candidate recall combination compared to the target recall combination includes the following steps: For each candidate recall combination, determine the first recommended product set corresponding to the target recall combination and its recommendation effectiveness score and recommendation computing cost index, as well as the second recommended product set corresponding to the candidate recall combination and its recommendation effectiveness score and recommendation computing cost index; Determining a recommendation effectiveness improvement degree according to the recommendation effectiveness score of the first recommended product set and the recommendation effectiveness score of the second recommended product set; Determine a difference lift according to the total number of different products between the first recommended product set and the second recommended product set and the total number of products in the first recommended product set; Determining a computing cost increase degree according to a recommended computing cost index of the first recommended product set and a recommended computing cost index of the second recommended product set; The recommendation value improvement is determined based on the recommendation effectiveness improvement, difference improvement and computing cost increase.

3. The product recall combination method according to claim 1, characterized in that: After obtaining the current target recall combination, the following steps are included: Respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, obtain the candidate product set recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product; Initializing and constructing the effectiveness distribution corresponding to each candidate product set according to the recommendation effectiveness scores of each recall source; According to the historical user behaviors of the target user and each user in the user group to which the target user belongs and corresponding to each candidate product set, the effectiveness distributions are updated accordingly; Randomly sampling the corresponding expected effectiveness probabilities for each effectiveness distribution, and adjusting the recall scores of the candidate products in each candidate product set according to the expected effectiveness probabilities to obtain a recommendation score; Based on the recommendation score sorting, multiple recommended products are selected and pushed to the target user.

4. The product recall combination method according to claim 1, characterized in that: Determining multiple recall sources associated with the recommendation scenario and their recommendation effectiveness scores includes the following steps: Based on each recall source associated with the recommendation scenario, recall corresponding target products for the user's product recommendation event, and push the target products recalled by each recall source to the user; Based on whether the user makes a conversion behavior on the target product after the target product is exposed to the corresponding user, the recommendation effectiveness score of each recall source is determined.

5. The product recall combination method according to claim 1, characterized in that: After obtaining the current target recall combination, the following steps are included: In response to the recommendation computing overhead overload event of the recommendation scenario, obtain the target recall combination corresponding to each independent station store serving the recommendation scenario; Determine whether each target recall combination meets preset similarity conditions, and group the target recall combinations that meet the preset similarity conditions into the same recall combination group; For each recall combination group, multiple target recall combinations with relatively low recommendation value enhancement and the highest target recall combination are screened out from the recall combination group, and the highest target recall combination is updated to the target recall combination corresponding to the independent station store serving the recommendation scenario corresponding to each of the lower target recall combinations.

6. The product recall combination method according to claim 1, characterized in that: After obtaining the current target recall combination, the following steps are included: Respond to the product recommendation event of the independent station store, determine the target recommendation scenario and target user corresponding to the event, obtain the candidate product set recalled by each recall source in the target recall combination corresponding to the scenario and the recall score of each candidate product; Determine the conversion value score of each recalled candidate product associated with the target user; The recommendation score is determined according to the recall score and the conversion value score of each candidate product, and a plurality of recommended products are selected based on the recommendation score sorting and pushed to the target user.

7. The product recall combination method according to claim 6, characterized in that: Before responding to the product recommendation event of the independent station store, the following steps are included: Obtain training samples and supervisory labels in a prepared training set, wherein the training samples include commodity feature information of commodities and user feature information of users, and the supervisory labels represent whether the user has made a conversion behavior on the commodity; Inputting the training sample into the conversion evaluation model to extract corresponding deep semantic features and obtain a conversion feature vector; Classify and map the conversion feature vectors to predict corresponding conversion value scores; The loss value corresponding to the conversion value score is calculated according to the supervision label, and the conversion evaluation model is updated according to the loss value until the model converges, so as to determine the conversion value score of the recalled product associated with the user who triggered the recall.

8. A product recall assembly device, characterized in that: include: The data construction module is used to determine multiple recall sources associated with the recommendation scenarios and their recommendation effectiveness scores for various preset recommendation scenarios corresponding to the independent station stores, add the recall source with the highest recommendation effectiveness score to the target recall combination, and add the remaining recall sources to the candidate recall set; An iteration start module is used to start the iteration process, combine each recall source in the candidate recall set with the target recall combination, obtain a corresponding plurality of candidate recall combinations, and determine the recommendation value improvement of each candidate recall combination compared with the target recall combination; The iteration continuation module is used to remove the combined recall source in the candidate recall combination corresponding to the highest recommendation value improvement from the candidate recall set when the highest recommendation value improvement is greater than or equal to a preset threshold, and update the candidate recall combination to the target recall combination to continue the iteration process; The iteration end module is used to end the iteration process when the highest recommendation value improvement is less than a preset threshold, and obtain the current target recall combination to prepare for the product recommendation serving the recommendation scenario.

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