Commodity Recall Optimization Method and Its Device, Equipment, Medium, and Product

Abstract

The present application relates to a method and apparatus, device, medium, and product for optimizing product recall. The method includes: obtaining a plurality of subsets of product data recalled by a plurality of preset recall sources, each subset including the recalled products and the matching degrees characterizing the recall of the products; determining a plurality of evaluation indicators for each recall source according to the user behavior data corresponding to the products in each subset of product data; respectively aggregating and determining the index scores of each recall source according to the respective evaluation indicators corresponding to each recall source; and combining the subsets of product data into a product recommendation set, where the products are sorted according to the actual scores of each product, and the actual score of each product is the product of the matching degree of the product and the index score of the recall source of the product. The present application can achieve a fine-grained quantification of the information contribution degree of the products obtained by different recall sources, so that a more accurate final recommendation result can be determined according to the actual scores of the products, and it is suitable for use on independent sites.

Description

Technical Field

[0001] This application relates to the field of e-commerce information technology, and particularly to a method for optimizing product recall, as well as a corresponding device, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] Recommending products on an e-commerce platform is a high-frequency application, widely serving application scenarios such as product search, product advertising placement, and product screening. It can improve the product matching efficiency, making it easier for merchants to sell their products and for consumers to meet their needs.

[0003] When applying a product recommendation algorithm, a key link is the product recall link. In the product recall link, multiple preset recall sources are activated to obtain corresponding products with different strategies or through different data source channels. The recalled products can be used as results according to the actual situation or can be further sorted and used.

[0004] The number of products obtained from each recall source is numerous, and the information contribution value corresponding to each product has not been uniformly quantified. Therefore, simply combining the products obtained from different recall sources as the final result often leads to the final obtained result products failing to meet expectations.

[0005] To address such problems, the traditional processing method is to simply adjust the weights of each recall source or match a neural network model based on deep learning to adjust the importance of products in different recall sources. However, since each recall source itself contains multiple products, and the number of products ultimately expected to be obtained is often small, this method has a very limited actual effect due to the rough information processing granularity. Moreover, when implementing it using a neural network model, its training cost is high, and it is difficult to achieve results for online stores deployed as independent sites.

[0006] In view of this, the applicant of this application has found a new way and explored a new idea corresponding to optimizing product recall, thus submitting this application. Summary of the Invention

[0007] The purpose of this application is to solve the above problems and provide a method for optimizing product recall, as well as a corresponding device, computer equipment, computer-readable storage medium, and computer program product.

[0008] To achieve the various purposes of this application, the following technical solutions are adopted:

[0009] On the one hand, to achieve one of the purposes of this application, a method for optimizing product recall is provided, including the following steps:

[0010] Obtain multiple subsets of product data recalled by multiple preset recall sources. Each subset contains the recalled products and the matching degrees representing the recall of the products.

[0011] Determine multiple evaluation metrics for each recall source based on the user behavior data corresponding to the products in each subset of product data.

[0012] Aggregate and determine the metric scores for each recall source respectively according to the respective evaluation metrics corresponding to each recall source.

[0013] Merge the subsets of product data into a product recommendation set, where the products are sorted according to their actual scores. The actual score of each product is the product of the matching degree of the product and the metric score of the recall source of the product.

[0014] Optionally, determining multiple evaluation metrics for each recall source based on the user behavior data corresponding to the products in each subset of product data includes the following steps:

[0015] Obtain the user behavior data corresponding to the products in the subset of product data corresponding to each recall source.

[0016] Determine the parameter values corresponding to the respective evaluation metrics for each recall source according to the user behavior data.

[0017] Apply the preset algorithms corresponding to the respective evaluation metrics and their corresponding parameter values to calculate the evaluation metrics corresponding to each recall source.

[0018] Optionally, aggregating and determining the metric scores for each recall source respectively according to the respective evaluation metrics corresponding to each recall source includes the following steps:

[0019] Sort each recall source according to each evaluation metric to obtain a list of sorted recall sources corresponding to each evaluation metric.

[0020] According to the numerical values of the evaluation metrics, apply a unified sorting score sequence to set the corresponding sorting scores from high to low for the recall sources in the list of sorted recall sources for each recall source.

[0021] Sum the sorting scores of the list of sorted recall sources corresponding to the respective evaluation metrics to obtain the metric score corresponding to each recall source.

[0022] Optionally, before the step of obtaining multiple subsets of product data recalled by multiple preset recall sources, the following steps are included:

[0023] In response to a product matching instruction, call multiple recall sources, and based on each recall source, match a corresponding subset of product data for the target product specified by the product matching instruction. The products in the subset of product data are feature-similar to the target product.

[0024] Optionally, after the step of combining the subset of commodity data into a commodity recommendation set, the following steps are included:

[0025] Perform reverse sorting on the commodity recommendation set according to the actual scores of the commodities in the commodity recommendation set;

[0026] Obtain a preset number of top-ranked commodities from the commodity recommendation set to form a commodity recommendation list;

[0027] Push the commodity recommendation list to the target terminal device.

[0028] Optionally, the evaluation indicators include any one or more of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve, and the harmonic score is determined according to the recall rate and the accuracy rate.

[0029] On the other hand, a commodity recall optimization device is provided to meet one of the purposes of this application, including a data acquisition module, an index determination module, an index summarization module, and a data merging module, where: the data acquisition module is used to acquire multiple subsets of commodity data recalled by a preset number of recall sources, and each subset contains the recalled commodity and the matching degree representing the recall of the commodity; the index determination module is used to determine multiple evaluation indicators of each recall source according to the user behavior data corresponding to the commodities in each subset of commodity data; the index summarization module is used to respectively summarize and determine the index scores of each recall source according to the respective evaluation indicators corresponding to each recall source; the data merging module is used to merge the subset of commodity data into a commodity recommendation set, where the sorting is based on the actual scores of each commodity, and the actual score of each commodity is the product of the matching degree of the commodity and the index score of the recall source of the commodity.

[0030] Optionally, the index determination module includes: a behavior acquisition unit, which is used to acquire the user behavior data corresponding to the commodities in the subset of commodity data corresponding to each recall source; a parameter determination unit, which is used to determine the parameter values corresponding to the respective evaluation indicators of each recall source according to the user behavior data; an index calculation unit, which is used to apply the preset algorithms corresponding to the respective evaluation indicators and their corresponding parameter values to calculate the evaluation indicators corresponding to each recall source.

[0031] Optionally, the metric summarization module includes: a sorting processing unit for sorting each recall source according to each evaluation metric to obtain a recall source sorting list corresponding to each evaluation metric; a sorting assignment unit for setting corresponding sorting scores from high to low for the recall sources in each recall source sorting list according to the numerical values of the evaluation metrics and applying a unified sorting score sequence; and a summation summarization unit for summing the sorting scores of the recall source sorting lists corresponding to each evaluation metric by matching preset weights to obtain the metric scores corresponding to each recall source.

[0032] Optionally, prior to the data acquisition module, it includes: a recall execution module for responding to a product matching instruction, invoking multiple recall sources, and matching corresponding subsets of product data for the target product specified by the product matching instruction based on each recall source, and the products in the subset of product data are feature-similar to the target product.

[0033] Optionally, after the data merging module, it includes: a merging and sorting module for reverse-sorting the product recommendation set according to the actual scores of the products in the product recommendation set; a list generation module for obtaining a preset number of products with higher rankings from the product recommendation set to form a product recommendation list; and a list pushing module for pushing the product recommendation list to the target terminal device.

[0034] Optionally, the evaluation metrics include any one or more of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve, and the harmonic score is determined according to the recall rate and the accuracy rate.

[0035] In another aspect, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, and 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 optimization method described in the present application.

[0036] In another aspect, a computer-readable storage medium provided to meet another purpose of the present application stores a computer program implemented according to the product recall optimization method in the form of computer-readable instructions, and when the computer program is called and run by a computer, it executes the steps included in the method.

[0037] In another aspect, a computer program product provided to meet another purpose of the present application includes a computer program / instructions, and when the computer program / instructions are executed by a processor, they implement the steps of the product recall optimization method described in any embodiment of the present application.

[0038] Compared with the prior art, the present application has multiple advantages, including at least: by determining multiple evaluation metrics for the subsets of product data obtained from multiple recall sources respectively, determining the corresponding metric scores for each recall source based on the multiple evaluation metrics of each recall source, and jointly determining the actual score of a product according to the matching degree between the metric scores and the product itself, the present application realizes a fine-grained quantification of the information contribution degree of the products obtained from different recall sources, enabling a more accurate final recommendation result to be determined based on the actual score of the product. The implementation process does not require the use of complex mathematical models, can be achieved with low system overhead, and has a low deployment cost, being particularly suitable for use in independent sites in e-commerce platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0040] Figure 1 is a schematic flowchart of a typical embodiment of the product recall optimization method of the present application.

[0041] Figure 2 is a schematic flowchart of the process of determining the respective evaluation metrics of each recall source in the embodiment of the present application.

[0042] Figure 3 is a schematic flowchart of the process of determining the metric scores of each recall source in the embodiment of the present application.

[0043] Figure 4 is a schematic flowchart of the process of obtaining a product recommendation list based on a product recommendation set in the embodiment of the present application.

[0044] Figure 5 is a principle block diagram of the product recall optimization device of the present application;

[0045] Figure 6 is a schematic structural diagram of a computer device adopted by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The embodiments of the present application are described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary only for explaining the present application and should not be construed as limiting the present application.

[0047] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present application means the presence of the stated features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any unit and all combinations of one or more of the associated listed items.

[0048] Those skilled in the art can understand that, unless otherwise defined, all terms used herein (including technical terms and scientific terms) have the same meaning as the general understanding of those of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0049] Those skilled in the art can understand that the "client", "terminal", and "terminal device" used herein include both devices with a wireless signal receiver that only has the ability to receive and no ability to transmit, and devices with receiving and transmitting hardware that have the receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices can include: cellular or other communication devices such as personal computers, tablet computers, etc., which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant), which can include a radio frequency receiver, pager, Internet / intranet access, web browser, notepad, calendar, and / or GPS (Global Positioning System) receiver; conventional laptop and / or palm-held computers or other devices, which are conventional laptop and / or palm-held computers or other devices with and / or including a radio frequency receiver. The "client", "terminal", and "terminal device" used herein can be portable, transportable, installed in a vehicle (air, sea, and / or land), or suitable for and / or configured to run locally, and / or run in a distributed form at any other location on the earth and / or in space. The "client", "terminal", and "terminal device" used herein can also be a communication terminal, an Internet access terminal, a music / video playback terminal, for example, it can be a PDA, MID (Mobile Internet Device), and / or a mobile phone with music / video playback function, or can also be a smart TV, a set-top box, etc.

[0050] The hardware referred to by names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer, and is a hardware device with the necessary components disclosed by the von Neumann principle, including a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit loads the program stored in the external memory into the memory for execution, executes the instructions in the program, and interacts with the input / output devices to complete specific functions.

[0051] It should be noted that the concept of "server" as referred to in this application can similarly be extended to apply to server clusters. According to the network deployment principles understood by those skilled in the art, the various servers should be logically divided. Physically, these servers can either be independent of each other but can be called through interfaces, or integrated into a single physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility and should not be restricted by this when implementing the network deployment method of this application.

[0052] One or several technical features of this application, unless expressly specified, can either be deployed on a server for implementation and accessed by a client remotely invoking the online service interface provided by the server, or directly deployed and run on the client for implementation and access.

[0053] The neural network models cited or possibly cited in this application, unless expressly specified, can either be deployed on a remote server and remotely invoked by the client, or deployed on a client with sufficient device capabilities for direct invocation. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0054] All kinds of data involved in this application, unless expressly specified, can either be remotely stored on a server or stored on a local terminal device, as long as it is suitable for being invoked by the technical solution of this application.

[0055] Those skilled in the art should be aware that although the various methods of this application are described based on the same concept and thus show commonality with each other, unless otherwise specified, these methods can all be executed independently. Similarly, for the various embodiments disclosed in this application, they are all proposed based on the same inventive concept. Therefore, concepts with the same expression, as well as concepts that are only appropriately transformed for convenience although the concept expressions are different, should be equivalently understood.

[0056] For the various embodiments to be disclosed in this application, unless expressly stated that there is a mutually exclusive relationship between them, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct new embodiments, as long as this combination does not deviate from the creative spirit of this application and can meet the requirements in the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0057] A method for optimizing product recall in this application can be programmed as a computer program product and implemented by running on a client or a server. For example, in an exemplary application scenario of this application, it can be implemented by deploying it on the server of an e-commerce platform. Thus, by accessing the interface opened after the computer program product runs, human-computer interaction can be carried out with the process of the computer program product through a graphical user interface to execute this method.

[0058] Please refer to Figure 1 , in a typical embodiment of the method for optimizing product recall in this application, the following steps are included:

[0059] Step S1100: Obtain multiple subsets of product data recalled by a preset multiple of recall sources, each subset containing the recalled products and the matching degrees characterizing the recall of these products.

[0060] In an e-commerce platform, multiple recall sources are configured. Each recall source provides a product data recall service corresponding to a recall strategy, matches the product data from its corresponding recall channel, and obtains the corresponding subset of product data. The subset of product data usually contains multiple products and the matching degrees of each product obtained under the corresponding recall source that match this recall source.

[0061] For example, in an application scenario of multi-channel recall based on a target product, a target product is given in advance, and multiple recall sources for implementing similarity matching are called. Among them, the first recall source performs similarity matching between the deep semantic feature information of the product picture and / or product description information of the target product and the deep semantic feature information of the product pictures and / or product description information of each product in the first product data pool, and correspondingly calculates the matching degrees of the products that match the target product, and then extracts the products with matching degrees higher than the preset threshold from the first product data pool to form the first subset of product data corresponding to the first recall source. The second recall source is the same as the first recall source. It performs similarity matching between the deep semantic feature information of the product picture and / or product description information of the target product and the deep semantic feature information of the product pictures and / or product description information of each product in the second product data pool, and correspondingly calculates the matching degrees of the products that match the target product, and then extracts the products with matching degrees higher than the preset threshold from the second product data pool to form the second subset of product data corresponding to the first recall source. The first product data pool can store high-profit products, and the second product data pool can store hot-selling products.

[0062] It is not difficult to understand from the above examples that those skilled in the art can preset the above-mentioned recall sources, preset corresponding recall strategies and / or recall channels for each recall source, implement corresponding recall services, provide corresponding recall interfaces. When it is necessary to recall commodity data, by calling the corresponding recall source, the corresponding subset of commodity data can be obtained. By concurrently calling multiple recall sources, multiple subsets of commodity data can be correspondingly obtained. Each subset of commodity data is obtained based on different recall strategies and / or recall channels. Therefore, there may be individual identical commodities among the commodities in the subsets of commodity data. For the identical commodities, the optimal selection can be made in the subsequent merging stage.

[0063] Step S1200: Determine multiple evaluation indicators for each recall source according to the user behavior data corresponding to the commodities in each subset of commodity data;

[0064] For each subset of commodity data obtained corresponding to each recall source, the practical effectiveness of each recall source can be analyzed through the user behavior data generated by each commodity being accessed by users.

[0065] The user behavior data can be obtained by embedding points in the commodity display page of a certain commodity. When any user accesses a commodity display page, a corresponding user behavior data is submitted to the server through the corresponding embedded point code of the page, representing that the user performs a corresponding operation on the corresponding commodity for the commodity display page and generates a corresponding operation event. The user behavior data is stored in the log database in the form of logs. According to the commodities recalled by each recall source, the user behavior data of the commodity within a preset time range can be directly called from the log database to determine the evaluation indicators of each recall source.

[0066] By statistically analyzing the user behavior data of the commodities in each subset of commodity data obtained by each recall source, various evaluation indicators corresponding to each recall source can be obtained. The evaluation indicators include any one or any combination of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve, which can be selected by those skilled in the art as needed. As an example, in this embodiment, the above-mentioned various evaluation indicators can be simultaneously used to evaluate the recall effectiveness of each recall source.

[0067] The so-called accuracy rate , in this application, is defined as the ratio of the number of commodities accessed by users in the subset of commodity data corresponding to each recall source to the total number of commodities in this subset of commodity data . The formula is expressed as follows:

[0068]

[0069] The so-called recall rate , in this application, is defined as the number of products accessed by users in the product data subset corresponding to each recall source Total number of products in each product data subset corresponding to all recall sources The formula is:

[0070]

[0071] The reconciliation score , is the harmonic mean of precision and recall, used to weigh the accuracy and recall. Generally speaking, accuracy and recall are negatively correlated. Generally speaking, there is a contradiction between precision and recall. Therefore, the influence of accuracy and recall can be balanced by introducing a harmonic score. The larger the harmonic score, the higher the model quality. The harmonic score can be calculated according to the following formula:

[0072]

[0073] in, It can be flexibly adjusted according to the importance of recall rate. When it is equal to 1, it means that accuracy and recall rate are equally important; when it is less than 1, it means that accuracy is more important than recall rate; when it is greater than 1, it means that recall rate is more important than accuracy. Since this application focuses on recall optimization, the exemplary setting is .

[0074] The area under the curve (AUC) is one of the indicators used to evaluate the quality of a recall source. In this application, it is defined by the following formula:

[0075]

[0076] in, is the number of products in the corresponding recall source that have not been visited by users. It can be seen that the area under the curve reflects the ranking ability of the overall samples recalled by the recall source.

[0077] The area under the user curve , is the result obtained by calculating the area under the curve for each user. In the field of computational advertising, what is actually measured is the ability of different users to rank different advertising products. Therefore, what should be paid more attention to is the ability of the same user to rank different advertisements. Therefore, the area under the curve can be calculated based on a single user. To this end, the following formula is used to determine:

[0078]

[0079] in Represents users and products

[0080] According to the above formula, it can be known that GAUC (group auc) actually calculates the Auc of each user, then performs weighted averaging, and finally obtains Gauc, which can reduce the influence that the ranking results among different users are not easy to compare. During actual processing, the weight can generally be set to the number of times each user accesses.

[0081] According to the disclosure of the above formulas for various exemplary evaluation metrics, it can be understood that when it is necessary to evaluate each recall source based on the corresponding subset of product data for each recall source, by using the relevant parameter values determined from the user behavior data and applying the corresponding evaluation metric formulas, the corresponding evaluation metrics can be calculated.

[0082] Step S1300: Aggregate and determine the metric scores for each recall source according to the respective evaluation metrics corresponding to each recall source;

[0083] For each recall source, its various evaluation metrics are relatively scattered, providing evaluation information about the corresponding recall source from different perspectives, which is not intuitive. Therefore, a preset aggregation algorithm can be applied to synthesize the multiple evaluation metrics corresponding to each recall source into a corresponding single metric score for comprehensively indicating the pros and cons of each recall source. In one embodiment, the above evaluation metrics are directly added to obtain the corresponding metric scores. In another embodiment, weights can also be assigned to the evaluation metrics for aggregation. For example, an exemplary formula is as follows:

[0084]

[0085]

[0086] where is the preset weight, which can be preset by those skilled in the art.

[0087] It is not difficult to understand that by aggregating the respective evaluation metrics corresponding to each recall source to determine the respective metric scores, the pros and cons of each recall source are comprehensively evaluated from different dimensions, unifying the overall performance of each recall source into the same dimension, which is convenient for guiding the subsequent sorting of all products recalled by all recall sources.

[0088] In addition to the above methods, other methods can also be used to determine the metric scores. The subsequent embodiments of this application will further disclose the implementation of other methods, which will not be elaborated here for the time being.

[0089] Step S1400: Merge the subset of product data into a product recommendation set, where the products are sorted according to their actual scores, and the actual score of each product is the product of the matching degree of the product and the metric score of the recall source of the product.

[0090] In order to summarize the products in the product data subsets corresponding to all recall sources, the respective product data subsets can be merged to obtain a product recommendation set. During the merging process, in the case where there are the same products in different product data subsets, only the data records corresponding to the recall sources with higher metric scores can be retained to achieve duplicate removal. The product recommendation set obtained through merging can be directly used as the result set, or can be pushed to a preset sorting model or apply a preset sorting algorithm, and then output after sorting, which can be implemented by those skilled in the art as needed.

[0091] To facilitate the unified sorting of the products in the product recommendation set, the matching degree obtained by each product in its corresponding recall source can be multiplied by the metric score obtained by the corresponding recall source to obtain the product, and used as the actual score corresponding to the product. Subsequently, the sorting of the product recommendation set can be achieved based on this actual score.

[0092] It is not difficult to understand from the above embodiments that, compared with the prior art, the present application has multiple advantages, including at least: the present application determines multiple evaluation metrics for the product data subsets obtained by multiple recall sources respectively, determines the metric score corresponding to each recall source according to the multiple evaluation metrics of each recall source, and jointly determines the actual score of the product based on the metric score and the matching degree of the product itself, realizing the fine-grained quantification of the information contribution degree of the products obtained by different recall sources, so that a more accurate final recommendation result can be determined according to the actual score of the product. The implementation process does not require the use of complex mathematical models, can be achieved with low system overhead, has a low deployment cost, and is particularly suitable for use in independent sites in e-commerce platforms.

[0093] Based on any of the above embodiments, please refer to Figure 2 , the step S1200, determining multiple evaluation metrics for each recall source according to the user behavior data corresponding to the products in each product data subset, includes the following steps:

[0094] Step S1210, obtaining the user behavior data corresponding to the products in the product data subset corresponding to each recall source;

[0095] As described above, for each recall source, the original user behavior data within a certain time range is obtained from the log database. Then, the original user behavior data is subjected to data cleaning so that the user behavior data can be formatted into a unified expression, and its form example is as follows:

[0096] User, current product, recalled product, matching degree, recall source, access status

[0097] Among them, the user instructs to execute operations on active users in the e-commerce platform. The current commodity indicates the target commodity used to trigger multiple recall sources for recall. The recalled commodity indicates the corresponding commodity recalled by the recall source according to the target commodity. The recall source indicates the recall source corresponding to the recalled commodity. The access status indicates whether the user accesses the recalled commodity, which is represented by a binary identifier. For example, if the user has accessed the recalled commodity, it is represented as 1; otherwise, it is represented as 0.

[0098] During the data cleaning process, any form of user access to the recalled commodity, such as clicking on the corresponding link, adding the recalled commodity to the shopping cart, placing an order for the recalled commodity, paying for the order corresponding to the recalled commodity, etc., can be normalized into a single access fact. That is, only whether the user has accessed the recalled commodity in any form is marked, regardless of the number of times the user has accessed. Accordingly, the relevant computational complexity and complexity are simplified, and the execution efficiency is improved.

[0099] Step S1220: Determine the parameter values corresponding to each evaluation index of each recall source according to the user behavior data;

[0100] To facilitate the application of the formulas corresponding to each evaluation index, it is necessary to determine the parameter values required for the parameters of each formula from the cleaned user behavior data based on each formula. Examining the exemplary formulas of the present application, it can be seen that the commodities in the commodity data subset of each recall source are divided into two categories: accessed and unaccessed. The accessed commodities are regarded as the positive samples of the recall source, and the unaccessed commodities are regarded as the negative samples of the recall source. Thus, the confusion matrix of each recall source can be obtained, and the corresponding parameter values can be statistically obtained according to this confusion matrix.

[0101] For example, according to the user behavior data after data cleaning, the commodities accessed by the user in the commodity data subset of each recall source are counted, and the number of accessed commodities can be obtained , and the number of unaccessed commodities for each recall source can also be correspondingly counted, thereby determining the parameter values corresponding to each evaluation index.

[0102] Step S1230: Apply the preset algorithms corresponding to each evaluation index and their corresponding parameter values to calculate the evaluation indexes corresponding to each recall source.

[0103] After determining the parameter values of each evaluation index, such as the number of accessed and unaccessed commodities, the evaluation index formulas of the present application that reference these parameter values can be applied to perform corresponding calculations on each evaluation index, and finally the corresponding result values of each evaluation index can be obtained.

[0104] It can be understood from the above embodiments that, based on user behavior data, the parameter values required for each evaluation index can be quickly counted, and the evaluation indexes of each recall source can be determined according to the user behavior data, so as to evaluate the advantages and disadvantages of each recall source with the same standard.

[0105] Based on any of the above embodiments, please refer to Figure 3 , the step S1300 of respectively aggregating and determining the index score of each recall source according to each evaluation index corresponding to each recall source includes the following steps:

[0106] Step S1310: Sort each recall source according to each evaluation index to obtain a recall source sorting list corresponding to each evaluation index;

[0107] For each evaluation index, each recall source can be sorted respectively, and a recall source sorting list can be obtained accordingly. For example, for the five evaluation indexes in the above example, five recall source sorting lists can be obtained accordingly. For the convenience of understanding, for each recall source sorting list here, each recall source is sorted from high to low according to its corresponding evaluation index.

[0108] Step S1320: According to the numerical value of the evaluation index, apply a unified sorting score sequence to set the corresponding sorting scores from high to low for the recall sources in each recall source sorting list;

[0109] A sorting score sequence is preset in advance. For example, it is assumed that there are N recall sources, and N is a natural number greater than 2. Then the sorting score sequence can be preset as follows:

[0110] [1, 2, …… N]

[0111] Then, for each recall source sorting list, adapt the sorting scores in the sorting score sequence from high to low according to the evaluation index. For example, the recall source with the highest evaluation index is matched with the sorting score N, the recall source with the lowest evaluation index is matched with the sorting score 1, and so on.

[0112] Step S1330: Sum the sorting scores of each recall source sorting list corresponding to each evaluation index to match the preset weights to obtain the index score corresponding to each recall source.

[0113] The sorting scores determined for each recall source through its corresponding evaluation indexes are used to convert the dimensions of different evaluation indexes to the same numerical space for representing sorting scores. Therefore, for each recall source corresponding to its different evaluation indexes with preset weights, the sorting scores of each recall source corresponding to different evaluation indexes can be weighted and summed, and the summation result can be used as the index score corresponding to each recall source.

[0114] For example, it is expressed by the following formula:

[0115]

[0116]

[0117] Among them, 、 、 、 、 are the sorting scores corresponding to the evaluation indicators such as the accuracy rate, recall rate, harmonic score, area under the curve, and user area under the curve respectively. is the recall source. is a preset weight, which can be preset by those skilled in the art. In an exemplary weight configuration scheme, the area under the curve and the user area under the curve can maintain the same highest weight, the recall rate and the harmonic score can maintain the same lowest weight, and the accuracy rate can be a compromise weight between the two.

[0118] According to the above embodiments, by using the sorting score as an intermediate dimension, the equivalent analysis corresponding to each evaluation indicator is realized, which facilitates the standardized adjustment of the roles of each evaluation indicator through the preset weight, thereby helping to adjust the finally obtained indicator score so as to quickly obtain an objective and effective evaluation result.

[0119] Optionally, before the step of obtaining multiple subsets of product data recalled by multiple preset recall sources, the following steps are included:

[0120] Respond to the product matching instruction, call multiple recall sources, and match a corresponding subset of product data for the target product specified by the product matching instruction based on each recall source, and the products in the subset of product data are feature-similar to the target product.

[0121] In an exemplary application scenario, the user triggers a product matching instruction in an online store on the independent site of the e-commerce platform and sends it to the server. The product matching instruction contains the specified target product to achieve similar product matching. Therefore, the server responds to the product matching instruction and enables the corresponding recall source to recall products, and accordingly obtains subsets of product data obtained by each recall source retrieving its similar products according to the target product.

[0122] In one embodiment, when each recall source retrieves similar products for a target product, it can calculate the data distance between the deep semantic information pre-extracted from the product information of the product and the deep semantic information pre-extracted from the products in the product data pool corresponding to the recall source, so as to determine the similarity between the two. Then, according to a preset threshold, products with a similarity higher than the preset threshold are retrieved from the product data pool to form a corresponding product data subset, which can be used as the recall result. In the product data subset, the similarity corresponding to each product can be associated and stored, and the similarity is used as the matching degree between the corresponding product and the target product.

[0123] It can be understood from the above embodiments that the present application can respond to user instructions through a pre-step to achieve user interaction, perform corresponding recalls according to the instructions of the terminal device, and then determine a product recommendation set that matches the target product on this basis.

[0124] Optionally, please refer to Figure 4 , after the step of merging the product data subsets into a product recommendation set, the following steps are included:

[0125] Step S1500: Perform reverse sorting on the product recommendation set according to the actual scores of the products in the product recommendation set;

[0126] As mentioned above, in the product recommendation set formed by merging the product data subsets of each recall source, each product has obtained its corresponding actual score. Since this actual score has undergone dimension conversion and weight matching based on the evaluation indicators of each recall source, it can effectively and uniformly represent the information contribution value of each product. Therefore, the products in the product recommendation set can be directly sorted in reverse according to this actual score to obtain a sorting result from large to small.

[0127] Step S1600: Obtain a preset number of products with higher rankings from the product recommendation set to form a product recommendation list;

[0128] Generally, for some specific application scenarios, such as product similarity matching, advertising product placement, etc., after obtaining a product recommendation set by recalling based on a target product, the number of products in the product recommendation set is relatively large, such as hundreds or thousands. However, in actual applications, usually only some products with higher information contribution values need to be displayed. Therefore, according to a preset number, the products in the sorted product recommendation set can be truncated, retaining a preset number of products with higher rankings and removing the other products behind, thereby obtaining a product recommendation list.

[0129] Step S1700: Push the product recommendation list to the target terminal device.

[0130] Finally, according to the actual scenario, push the product recommendation list to the terminal device that drives product recall, such as the terminal device of the user who provides the target product, so that the product recommendation list can be parsed and displayed in the graphical user interface of the terminal device.

[0131] As can be seen from the above embodiments, the present application can quickly obtain a product recommendation list with a relatively low computational amount, which can meet requirements such as product similarity matching and advertising product placement, and is particularly suitable for being deployed in an online store of an e-commerce platform implemented based on an independent site. Its deployment cost is low, but the obtained recall results are accurate and efficient.

[0132] Please refer to Figure 5 , to provide a product recall optimization device for one of the purposes of the present application, which is a functional embodiment of the product recall optimization method of the present application. The device includes a data acquisition module 1100, an index determination module 1200, an index summary module 1300, and a data merging module 1400, where: the data acquisition module 1100 is used to acquire multiple product data subsets recalled by a preset multiple recall sources, and each subset contains the recalled product and the matching degree representing the recall of the product; the index determination module 1200 is used to determine multiple evaluation indexes of each recall source according to the user behavior data corresponding to the products in each product data subset; the index summary module 1300 is used to respectively summarize and determine the index scores of each recall source according to the respective evaluation indexes corresponding to each recall source; the data merging module 1400 is used to merge the product data subsets into a product recommendation set, where the actual scores of each product are sorted according to the actual scores of each product, and the actual score of each product is the product of the matching degree of the product and the index score of the recall source of the product.

[0133] Optionally, the index determination module 1200 includes: a behavior acquisition unit, configured to acquire the user behavior data corresponding to the products in the product data subset corresponding to each recall source; a parameter determination unit, configured to determine the parameter values corresponding to each evaluation index of each recall source according to the user behavior data; an index calculation unit, configured to calculate the evaluation indexes corresponding to each recall source by applying the preset algorithms corresponding to each evaluation index and their corresponding parameter values.

[0134] Optionally, the index summary module 1300 includes: a sorting processing unit, configured to sort each recall source according to each evaluation index to obtain a recall source sorting list corresponding to each evaluation index; a sorting assignment unit, configured to set corresponding sorting scores from high to low for the recall sources in each recall source sorting list according to the numerical values of the evaluation indexes by applying a unified sorting score sequence; a summation summary unit, configured to sum the sorting scores of the recall source sorting lists corresponding to each evaluation index by matching preset weights to obtain the index score corresponding to each recall source.

[0135] Optionally, prior to the data acquisition module 1100, it includes: a recall execution module, which is used to respond to a product matching instruction, call multiple recall sources, and match corresponding subsets of product data for the target product specified by the product matching instruction based on each recall source, and the products in the subset of product data are feature-similar to the target product.

[0136] Optionally, after the data merging module 1400, it includes: a merging and sorting module, which is used to perform reverse sorting on the product recommendation set according to the actual scores of the products in the product recommendation set; a list generation module, which is used to obtain a preset number of products with higher rankings from the product recommendation set to form a product recommendation list; a list pushing module, which is used to push the product recommendation list to the target terminal device.

[0137] Optionally, the evaluation metrics include any one or more of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve, and the harmonic score is determined according to the recall rate and the accuracy rate.

[0138] To solve the above technical problems, an embodiment of the present application also provides a computer device. As Figure 6 shown, it is a schematic internal structure diagram of the computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected through 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 control information sequence can be stored in the database. When the computer-readable instructions are executed by the processor, the processor can implement a product search category recognition 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 computer-readable instructions can be stored in the memory of the computer device. When the computer-readable instructions are executed by the processor, the processor can execute the product recall optimization method of the present application. The network interface of the computer device is used to communicate with the terminal. Those skilled in the art can understand that Figure 6 the structure shown in

[0139] In this embodiment, the processor is used to execute Figure 5The specific functions of each module and its sub-modules therein, and the memory stores the program codes and various types of data required to execute the above-mentioned modules or sub-modules. The network interface is used for data transmission between user terminals or servers. In this embodiment, the memory stores the program codes and data required to execute all modules / sub-modules in the product recall optimization device of the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

[0140] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to execute the steps of the product recall optimization method according to any embodiment of the present application.

[0141] The present application also provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by one or more processors, the steps of the method according to any embodiment of the present application are implemented.

[0142] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. 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 methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0143] In summary, the present application realizes the fine-grained quantification of the information contribution degree of the products obtained from different recall sources, so that more accurate final recommendation results can be determined according to the actual scores of the products. The implementation process does not require the use of complex mathematical models, can be realized with low system overhead, has a low deployment cost, and is particularly suitable for use in independent sites in e-commerce platforms.

[0144] Those skilled in the art of the present technology can understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, the other steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and solutions in the prior art that are the same as those disclosed in the various operations, methods, and processes in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0145] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for optimizing product recall, characterized in that, Including the following steps: Obtain multiple subsets of product data recalled by a preset plurality of recall sources, each subset including the recalled products and the matching degrees representing the recall of the products; Determine multiple evaluation indicators for each recall source according to the user behavior data corresponding to the products in each subset of product data, including: obtaining the user behavior data corresponding to the products in the subset of product data corresponding to each recall source; determining the parameter values corresponding to each evaluation indicator of each recall source according to the user behavior data; applying the preset algorithms corresponding to each evaluation indicator and their corresponding parameter values to calculate the evaluation indicators corresponding to each recall source; Summarize and determine the indicator scores for each recall source according to the respective evaluation indicators corresponding to each recall source, including: sorting each recall source according to each evaluation indicator to obtain a list of sorted recall sources corresponding to each evaluation indicator; applying a unified sequence of sorting scores according to the numerical values of the evaluation indicators to set corresponding sorting scores from high to low for the recall sources in the list of sorted recall sources for each recall source; summing the sorting scores of the list of sorted recall sources corresponding to each evaluation indicator to obtain the indicator score corresponding to each recall source; Merge the subsets of product data into a product recommendation set, where the product recommendation set is sorted according to the actual scores of each product, and the actual score of each product is the product of the matching degree of the product and the indicator score of the recall source of the product.

2. The optimized method for product recall according to claim 1, characterized in that, Before the step of obtaining multiple subsets of product data recalled by a preset plurality of recall sources, including the following steps: In response to a product matching instruction, call a plurality of recall sources, and based on each recall source, match a corresponding subset of product data for the target product specified by the product matching instruction, and the products in the subset of product data are feature-similar to the target product.

3. The optimized method for product recall according to claim 1, wherein, After the step of merging the subsets of product data into a product recommendation set, including the following steps: Reverse-sort the product recommendation set according to the actual scores of the products in the product recommendation set; Obtain a preset plurality of products ranked at the top from the product recommendation set to form a product recommendation list; Push the product recommendation list to the target terminal device.

4. The optimized method for product recall according to any one of claims 1 to 3, characterized in that, The evaluation indicators include any one or any combination of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve. The harmonic score is determined according to the recall rate and the accuracy rate. The user area under the curve is the result obtained by calculating the area under the curve based on each user.

5. An optimized device for product recall, characterized in that, Including: A data acquisition module for obtaining multiple subsets of product data recalled by a preset plurality of recall sources, each subset including the recalled products and the matching degrees representing the recall of the products; An indicator determination module for determining multiple evaluation indicators for each recall source according to the user behavior data corresponding to the products in each subset of product data, including: obtaining the user behavior data corresponding to the products in the subset of product data corresponding to each recall source; determining the parameter values corresponding to each evaluation indicator of each recall source according to the user behavior data; applying the preset algorithms corresponding to each evaluation indicator and their corresponding parameter values to calculate the evaluation indicators corresponding to each recall source; An index summarization module, configured to respectively summarize and determine the index scores of each recall source according to the respective evaluation indexes corresponding to each recall source, including: sorting each recall source according to each evaluation index respectively to obtain a recall source sorting list corresponding to each evaluation index; according to the numerical value of the evaluation index, applying a unified sorting score sequence to set corresponding sorting scores from high to low for the recall sources in each recall source sorting list; summing the sorting scores of the recall source sorting lists corresponding to each evaluation index by matching with a preset weight to obtain the index score corresponding to each recall source; A data merging module, configured to merge the subset of commodity data into a commodity recommendation set, wherein the commodity recommendation set is sorted according to the actual score of each commodity, and the actual score of each commodity is the product of the matching degree of the commodity and the index score of the recall source of the commodity.

6. The optimized device for product recall according to claim 5, characterized in that Subsequent to the data merging module, the apparatus further includes: A merging and sorting module, configured to perform reverse sorting on the commodity recommendation set according to the actual scores of the commodities in the commodity recommendation set; A list generation module, configured to obtain a preset number of commodities with higher rankings from the commodity recommendation set to form a commodity recommendation list; A list pushing module, configured to push the commodity recommendation list to a target terminal device.

7. The optimized device for product recall according to claim 5 or 6, characterized in that, The evaluation index includes any one or more of the following: recall rate, accuracy rate, harmonic score, area under the curve, user area under the curve, the harmonic score is determined according to the recall rate and the accuracy rate, and the user area under the curve is a result obtained by calculating the area under the curve based on each user.

8. A computer device, comprising a central processing unit and a memory, characterized in that, The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium, characterized in that, It stores a computer program implemented according to the method according to any one of claims 1 to 4 in the form of computer-readable instructions, and when the computer program is called and run by a computer, it executes the steps included in the corresponding method.

10. A computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images