CN114663155A - Advertisement putting and selecting method and device, equipment, medium and product thereof - Google Patents

Abstract

The application discloses a method for advertisement putting and selecting, a device, equipment, a medium and a product thereof, wherein the method comprises the following steps: acquiring advertisement prediction data corresponding to each commodity in a commodity database of an independent site, wherein the advertisement prediction data comprise an acquisition rate and effect data, the acquisition rate represents the prediction probability of the commodity for advertisement delivery, and the effect data comprise a conversion probability representing possible harvest of the commodity after advertisement delivery; calculating a recommendation score corresponding to each commodity according to the acquisition rate and the effect data of the commodity; and selecting commodities in the commodity database as advertisement putting options according to the recommendation scores to construct an advertisement putting option recommendation list. The method and the device can accurately match the advertisement putting selections which accord with the preference of the user, and can obtain good advertisement putting effect.

Description

Advertising selection method and device, equipment, medium and product

technical field

The present application relates to the field of e-commerce information technology, and in particular, to a method for selecting products for advertisement placement and a corresponding device, computer equipment, computer-readable storage medium, and computer program products.

Background technique

In the cross-border e-commerce service platform, it provides basic technical services for a large number of independent sites, and each independent site usually has an independent domain name. To recommend products, it is necessary to be able to select products on its platform and determine products with sales potential that are suitable for advertising, so as to promote terminal sales or configuration of products on the shelves.

For the sales potential of commodities, management users of independent sites generally have relevant market trends to consider, such as newly listed commodities, commodities of a certain category, commodities with low prices, etc. The corresponding market trends are good and they have certain market competitiveness. Sales potential and market trends need to be analyzed with the help of abundant commodity advertising data. However, at this stage, commodity advertising data is sparsely distributed. Due to the limited financial resources of most independent sites, the investment in commodity advertising is limited or even very little. , therefore, there are only a few independent sites with abundant commodity advertising data, while the commodity advertising data of most independent sites is scarce. More importantly, these scarce commodity advertising data are limited by the limited input cost and are underexposed. , making the corresponding returns difficult to reflect.

The scarcity of data and the low reference value of the data make it difficult for the management users of most independent sites to determine the commodities with sales potential. So far, there is no better solution for this in the prior art.

In order to solve the deficiencies of the prior art, the applicant has made corresponding explorations.

SUMMARY OF THE INVENTION

The primary purpose of the present application is to solve at least one of the above problems and provide a method for selecting advertisements and its corresponding apparatus, computer equipment, computer-readable storage medium, and computer program product.

In order to meet the various purposes of the application, the application adopts the following technical solutions:

A method for selecting products for advertisement placement provided in accordance with one of the purposes of this application includes the following steps:

Obtain the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the advertisement prediction data includes the adoption rate and the effectiveness data, the adoption rate represents the predicted probability of the commodity being advertised, and the effectiveness data includes the characteristics representing the commodity The conversion probability that may be harvested after being placed in the advertisement;

According to the adoption rate and effectiveness data of the product, the recommendation score corresponding to each product is calculated;

According to the recommendation score, commodities in the commodity database are selected as advertisement placement options to construct an advertisement placement option recommendation list.

In a further embodiment, before the step of acquiring the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the following steps are included:

Determine the adoption rate of each commodity in the commodity database of the independent site using an adoption rate prediction model pre-trained to a convergent state;

Determine the click-through rate of each item in the item database of the independent site using a CTR prediction model pre-trained to a convergent state;

Determine the conversion rate of each item in the item database of the independent site using a conversion rate prediction model pre-trained to a convergent state;

The adoption rate and effectiveness data of each commodity are stored in the commodity database corresponding to each commodity, and the effectiveness data is the product of the click-through rate and the conversion rate associated with the same commodity.

In a further embodiment, in the step of determining the adoption rate of each commodity in the commodity database of the independent site by using the adoption rate prediction model trained in advance to a convergent state, the training process of the adoption rate prediction model includes the following steps:

A training sample is called from the data set. The training sample includes positive samples and negative samples. Each training sample is provided with a supervision label indicating whether the product is advertised. The product information and statistical information of the advertised products, the negative sample includes the product information and statistical information of the unadvertised products in the product database of the e-commerce platform, and the product information includes the product shelf time, product category, and product price. The above statistical information includes click rate, purchase rate, repurchase rate;

Inputting the training sample into the adoption rate prediction model to obtain a commodity feature vector and a statistical feature vector corresponding to the commodity information and statistical information in the training sample;

Splicing the commodity feature vector and the statistical feature vector to obtain a feature fusion vector, and mapping its input fully connected layer to a preset classification space to obtain a prediction adoption rate;

Calculate the loss value corresponding to the cross-entropy loss of the predicted adoption rate according to the supervision label, and judge whether the loss value reaches the preset threshold. When it reaches the preset threshold, the training is terminated; otherwise, the training is terminated according to the loss value. The model implements gradient update, and the next training sample in the data set is called to continue the iterative training of the model.

In a further embodiment, in the step of determining the click-through rate of each commodity in the commodity database of an independent site by using a click-through rate prediction model pre-trained to a convergent state, the training process of the click-through rate prediction model includes the following iterative steps: :

Acquiring training samples in the data set, where the training samples include commodity information of commodities that have been advertised and clicked by the user in the commodity database of the e-commerce platform, and a supervision label indicating whether the advertisement of the commodity has been clicked by the user;

Inputting the training sample into the click-through rate prediction model to extract corresponding deep semantic features to obtain feature vectors;

Use a classifier to classify and map the feature vector, and predict the corresponding click-through rate;

A cross-entropy loss value corresponding to the click-through rate predicted by the classifier is calculated according to the supervision label, and a gradient update is performed on the model according to the cross-entropy loss value until the model converges.

In a further embodiment, in the step of determining the conversion rate of each commodity in the commodity database of the independent site by using a conversion rate prediction model that has been pre-trained to a convergent state, the training process of the conversion rate prediction model includes the following iterative steps. :

Obtain training samples in the data set, the training samples include commodity information of commodities that have been advertised in the commodity database of the e-commerce platform and have been clicked by users to generate relevant user behaviors, and whether the advertisements of the commodities are generated after being clicked by users Supervision labels corresponding to relevant user behaviors;

Inputting the training sample into the conversion rate prediction model to extract corresponding deep semantic features to obtain feature vectors;

Use a classifier to classify and map the feature vector to predict the corresponding conversion rate;

A cross-entropy loss value corresponding to the conversion rate predicted by the classifier is calculated according to the supervision label, and a gradient update is performed on the model according to the cross-entropy loss value until the model converges.

In an extended embodiment, selecting a commodity in the commodity database as an advertisement placement selection according to the recommendation score to construct an advertisement placement selection recommendation list, including the following steps:

Sort according to the recommendation scores corresponding to the commodities in the commodity database, and construct a recommendation list for advertisement placement;

In response to an advertisement recommendation request from an online store, the advertisement placement recommendation list is pushed to the management user of the independent site.

A device for selecting products for advertisement placement provided for one of the purposes of this application includes: a data acquisition module, a score calculation module, and a list construction module, wherein the data acquisition module is used to acquire each commodity in the commodity database of an independent site Corresponding advertisement prediction data, the advertisement prediction data includes adoption rate and effectiveness data, the adoption rate represents the predicted probability of the product being advertised, and the effectiveness data includes the conversion probability representing the possible harvest after the product is advertised The score calculation module is used to calculate the corresponding recommendation score of each product according to the adoption rate and effectiveness data of the product; the list building module is used to select the product in the product database as an advertisement according to the recommendation score. The selection builds a recommended list of advertisement placement selections.

In a further embodiment, before the data acquisition module, it includes: an adoption rate determination module for determining the adoption rate of each commodity in the commodity database of the independent site by using a pre-trained adoption rate prediction model to a convergent state; click-through rate confirmation The module is used to determine the click rate of each commodity in the commodity database of the independent site by using the click rate prediction model pre-trained to the convergence state; the conversion rate confirmation module is used to determine the independent site using the conversion rate prediction model pre-trained to the convergence state The conversion rate of each commodity in the commodity database; the data storage sub-module is used to store the adoption rate of each commodity and the effectiveness data corresponding to each commodity in the commodity database, and the effectiveness data is related to the same commodity. The product of the click-through rate and the conversion rate.

In a further embodiment, in the adoption rate determination module, the training process of the adoption rate prediction model includes: a training sample invoking module for invoking a training sample from the data set, and the training sample includes a positive sample and a Negative samples, each training sample is provided with a supervision label indicating whether the product is advertised, the positive sample includes the product information and statistical information of the advertised products in the product database of the e-commerce platform, and the negative sample includes the electronic product. The commodity information and statistical information of the unadvertised commodities in the commodity database of the merchant platform, the commodity information includes the commodity shelf time, commodity category, commodity price, and the statistical information includes the click rate, the purchase rate, and the repurchase rate; feature extraction module for inputting the training sample into the adoption rate prediction model to obtain commodity feature vectors and statistical feature vectors corresponding to the commodity information and statistical information in the training sample; a vector splicing module for splicing the commodity features vector and the statistical feature vector to obtain a feature fusion vector, and map its input fully connected layer to a preset classification space to obtain a prediction adoption rate; an iterative training module is used to calculate the crossover of the prediction adoption rate according to the supervision label. The loss value corresponding to the entropy loss, determine whether the loss value reaches the preset threshold, and when it reaches the preset threshold, the training is terminated; otherwise, the gradient update is performed on the model according to the loss value, and the next training in the data set is called. The samples continue to perform iterative training on the model.

In a further embodiment, in the click-through rate confirmation module, the iterative execution training process of the click-through rate prediction model includes: a training sample calling module for acquiring training samples in the data set, and the training samples include e-commerce platforms. The product information of the product that has been advertised and clicked by the user in the product database of the product, and the corresponding supervision label indicating whether the advertisement of the product has been clicked by the user; the feature extraction module is used to input the training sample into the click-through rate. The prediction model extracts corresponding deep semantic features to obtain feature vectors; the classification mapping module is used to classify and map the feature vectors by using a classifier, and predict the corresponding click rate; the gradient update module is used for according to the supervision label. A cross-entropy loss value corresponding to the click-through rate predicted by the classifier is calculated, and a gradient update is performed on the model according to the cross-entropy loss value until the model converges.

In a further embodiment, in the conversion rate confirmation module, the iterative execution training process of the conversion rate prediction model includes: a training sample calling module for acquiring training samples in the data set, and the training samples include e-commerce platforms. Commodity information of commodities that have been advertised and clicked by users to generate relevant user behaviors in the commodity database, and a supervision label indicating whether the advertisements of the commodity have been clicked by users to generate corresponding user behaviors; feature extraction module, used for Inputting the training sample into the conversion rate prediction model to extract corresponding deep semantic features to obtain a feature vector; a classification mapping module is used to classify and map the feature vector by using a classifier to predict the corresponding conversion rate; A gradient update module, configured to calculate a cross-entropy loss value corresponding to the conversion rate predicted by the classifier according to the supervision label, and perform gradient update on the model according to the cross-entropy loss value until the model converges.

In an extended embodiment, the list building module includes: a scoring and sorting module, used for sorting according to the recommendation scores corresponding to the products in the product database, and constructing a list of recommended products for advertisement placement; a list push module, used for responding For an advertisement recommendation request of an online store, push the advertisement placement recommendation list to the management user of the independent site.

A kind of computer equipment that is adapted to 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 advertisement placement options described in the present application steps of the method.

A computer-readable storage medium provided for another purpose of the present application, which stores a computer program implemented according to the described method for placing advertisements in the form of computer-readable instructions, and the computer program is invoked by a computer to run , perform the steps included in the method.

A computer program product provided in accordance with another object of the present application includes a computer program/instruction, when the computer program/instruction is executed by a processor, the steps of the method described in any one of the embodiments of the present application are implemented.

It can be known from the typical embodiments of the present application and the modified embodiments thereof that the technical solutions of the present application have many advantages, including but not limited to the following aspects:

First, the present application calculates the corresponding recommendation score by obtaining the adoption rate and effectiveness data of the predicted product, and then constructs a recommended list of advertisement placement products according to the recommendation score and pushes it to the management user of the independent site. The difference from the prior art is only Relying on the performance data of the predicted products as the reference index for advertising product selection, the technical solution here integrates the product’s corresponding adoption rate and the performance data of the two dimensions of the recommendation score obtained as the reference index for advertising product selection, which can accurately predict the product selection. The probability of placing advertisements, and predicting the probability of conversions that may be harvested after the products are advertised, reflect the degree of suitability of the products for advertisements more comprehensively with quantified values, which is convenient for the management of independent sites. Users can quickly make decisions based on the recommendation score, and select the best choice. Products that are suitable for advertising and can perform well after advertising.

Secondly, on the one hand, most independent sites have limited financial resources and are unable to have most of the rich product advertisement data, resulting in the inability to decide suitable products for advertisement. The advertisement recommendation list can effectively assist this type of advertisement The management user of the independent site selects the products for advertisement placement on the products of the independent site, so as to solve the real pain point of the management user; The recommendation list can more effectively assist such novice users to select suitable products on their own independent sites, so as to place advertisements on them, improve user experience, and increase user stickiness.

In addition, the technical solution of the present application is simple and easy to implement, the time complexity of model calculation is low, and the engineering can be parallelized on a large scale.

Description of drawings

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

1 is a schematic flowchart of a typical embodiment of the method for selecting products for advertisement placement of the application;

2 is a schematic flowchart of the adoption rate and effectiveness data of pre-built commodities in the embodiment of the application;

3 is a schematic flowchart of the training process of the adoption rate prediction model described in the embodiment of the application;

4 is a schematic flowchart of a training process of the click-through rate prediction model described in the embodiment of the application;

5 is a schematic flowchart of the training process of the conversion rate prediction model described in the embodiment of the application;

FIG. 6 is a schematic flowchart of constructing a recommendation list for advertisement placement selection for pushing in an embodiment of the present application;

Fig. 7 is the principle block diagram of the advertisement placing product selection device of the present application;

FIG. 8 is a schematic structural diagram of a computer device used in this application.

Detailed ways

The following describes in detail the embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, but not to be construed as a limitation on the present application.

It will be understood by those skilled in the art that the singular forms "a", "an", "the" and "the" as used herein can include the plural forms as well, unless expressly stated otherwise. It should be further understood that the word "comprising" used in the specification of this application refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not preclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that when we refer to an element as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Furthermore, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combination of one or more of the associated listed items.

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 commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms, such as those defined in a general dictionary, should be understood to have meanings consistent with their meanings in the context of the prior art and, unless specifically defined as herein, should not be interpreted in idealistic or overly formal meaning to explain.

Those skilled in the art can understand that the "client", "terminal" and "terminal device" used herein include both a wireless signal receiver device that only has a wireless signal receiver without transmission capability, and a wireless signal receiver device. A device with receive and transmit hardware that has receive and transmit hardware capable of two-way communication over a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, 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, Personal Communications System) ), which can combine voice, data processing, fax and/or data communication capabilities; PDA (Personal Digital Assistant), which can include radio frequency receivers, pagers, Internet/Intranet access, web browsers, notepads , calendar and/or GPS (Global Positioning System) receivers; conventional laptop and/or palmtop computers or other devices having and/or conventional laptop and/or palmtop radio frequency receivers computer or other device. As used herein, "client", "terminal", "terminal device" may be portable, transportable, mounted in a vehicle (air, marine and/or land), or adapted and/or configured to be locally operate, and/or in distributed form, operate at any other location on Earth and/or space. The "client", "terminal" and "terminal device" used here can also be a communication terminal, an Internet terminal, and a music/video playing terminal, such as a PDA, MID (Mobile Internet Device) and/or A mobile phone with music/video playback function, or a smart TV, set-top box, etc.

The hardware referred to by names such as "server", "client" and "service node" in this application is essentially an electronic device with the equivalent capability of a personal computer, which is a central processing unit (including an arithmetic unit and a controller). ), memory, input device and output device and other necessary components disclosed by the Von Neumann principle, the computer program is stored in its memory, and the central processing unit transfers the program stored in the external memory into the memory to run, and executes the program. The instructions in the interface interact with input and output devices to complete specific functions.

It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters in the same way. According to the principles of network deployment understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent from each other but can be called through interfaces, or can be integrated into a physical server. A computer or a group of computers. Those skilled in the art should understand this modification, but should not limit the implementation of the network deployment manner of the present application.

Unless explicitly specified, one or more technical features of the present application can be deployed on the server and remotely invoked by the client to obtain the online service interface provided by the server to implement access, or can be directly deployed and run on the client to implement access.

All kinds of data involved in this application, unless specified in plain text, can be stored in a server remotely or in a local terminal device, as long as it is suitable for being called by the technical solution of this application.

Those skilled in the art should know that: although the various methods of the present application are described based on the same concept to show commonality with each other, unless otherwise specified, these methods can be independently executed. Similarly, for the various embodiments disclosed in this application, they are all proposed based on the same inventive concept. Therefore, the concepts expressed in the same way, and the concepts that are appropriately transformed for convenience even though the concept expressions are different, should be regarded as equivalent. understand.

In the various embodiments to be disclosed in this application, unless the mutually exclusive relationship between each other is clearly indicated, the related technical features involved in the various embodiments can be cross-combined to flexibly construct new embodiments, as long as the combination does not deviate from the present invention. The creative spirit of the application can meet the needs in the prior art or solve a certain aspect of the deficiencies in the prior art. Variations on this will be known to those skilled in the art.

A method for selecting products for advertisement placement of the present application can be programmed as a computer program product, which can be implemented by being deployed in a client or a server. The method can be performed by accessing the interface opened after the computer program product runs, and performing human-computer interaction with the process of the computer program product through a graphical user interface.

Please refer to FIG. 1 , in a typical embodiment of the method for selecting products for advertisement placement of the present application, the method includes the following steps:

Step S1100: Obtain the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the advertisement prediction data includes adoption rate and effect data, the adoption rate represents the predicted probability that the product is advertised, and the effect data includes Indicates the conversion probability that the product may be harvested after being advertised;

The technical solution of the present application takes the operating environment of an e-commerce service platform as its application environment, and the e-commerce service platform may be an e-commerce service platform that opens independent site services, typically, such as a cross-border e-commerce service platform. Since such platforms need to consider the network environment between different parts of the world and the independence of each merchant, by configuring each merchant's store as an independent site, the e-commerce service platform can serve a large number of such platforms. independent site.

Each of the independent sites has a commodity database corresponding to the commodities on sale on its own website. The commodity database contains a large amount of commodity information used to describe each commodity for sale. The commodity information includes but is not limited to commodity pictures, Various types of data such as product title, product details, product price, product sku (category), inventory, and product launch time. In addition, in order to sell more products on the shelves, the management users of independent sites will be on the e-commerce service platform. Advertisements are placed on products on the shelves and on sale, so that the products on the shelves and on sale are exposed to users, and before the execution process of advertisement placement on the products on the shelves and on sale, the advertisement prediction data is pre-generated to assist the management of the independent site. The products on the shelf are optimized and then the advertisement is placed. Therefore, the advertisement prediction data is associated with the corresponding products on the shelf and stored in the product database. Based on this, it is not difficult to understand that when the management user of the independent site triggers the advertisement prediction data acquisition request, the response to the request can be provided by invoking the e-commerce service platform to provide the corresponding data access interface to the commodity database of each independent site. Therefore, the access operation is performed on the commodity database of the independent site, and the advertisement prediction data corresponding to each commodity in the commodity database can be obtained.

The advertising effectiveness data includes adoption rate and effectiveness data, and the adoption rate represents the predicted probability of the product being advertised. features, and the adoption rate is high if it has the features suitable for advertising, otherwise the adoption rate is low if it does not have the feature. The features include product features and statistical features. Hot-selling commodity categories and low commodity prices in the commodity sales market. The statistical features include high click-through rate, high repurchase rate, high sales volume, and high purchase rate after the product is advertised. It can be seen that this feature is consistent with the management of independent sites. The market trend related to the selected product is considered when launching the product selection. Therefore, the product with this characteristic has a certain market competitiveness, that is, the sales potential.

The performance data includes the conversion probability representing the possible harvest after the product is advertised, the conversion probability includes predicting the click rate corresponding to the user's click after the product is advertised, and predicting that after the product is advertised and clicked by the user, other users will be generated. The conversion rate corresponding to the behavior. The other user behaviors include but are not limited to adding to the shopping cart, purchasing, repurchasing and other behaviors. It is not difficult to understand that the effect data reflects the effect obtained after the product is advertised. High efficiency means that the product can achieve good results after being advertised. Otherwise, it may achieve poor results. The good results can reflect the user's interest in the advertised product and even purchase it in need.

Step S1200, calculating the recommendation score corresponding to each product according to the adoption rate and effectiveness data of the product;

Since the adoption rate and performance data of the commodities are reference indicators that can assist the management users of independent sites to select commodities to be used for advertising, accordingly, in order to obtain a more comprehensive and intuitive numerical representation, the number of reference indicators is reduced. The reference index combining the two dimensions of the current adoption rate and the effectiveness data is reduced to a single reference index represented by the numerical value. The value is obtained by multiplying the values, and the value range is 0-100%, so that the value is used as the recommendation score. For example, the purchase rate is 80%, the click rate is 85%, and the conversion rate is 95%. Multiplying the corresponding recommendation score is 64.5%. It is not difficult to understand that the higher the recommendation score is, the higher the sales potential of the product is, and the better results are obtained after the product is advertised.

Step S1300 , selecting commodities in the commodity database as advertisement placement options according to the recommendation score to construct a recommendation list of advertisement placement options.

Further, according to the recommendation scores corresponding to each commodity in the commodity database, select a plurality of commodities with the highest recommendation scores as the advertisement placement selections, and compare the recommended scores corresponding to the advertisement placement selections in descending order. The advertisement placement selections are sorted, and then the advertisement placement selections under the sorting are constructed into an advertisement placement selection recommendation list, and the specific number of selected products can be selected by those skilled in the art according to the actual situation. The advertisement placement selection recommendation list The number of displayed products is allowed to be flexible, so that the recommended list of products for advertisement placement is pushed to the administrator user of the independent site, so that the administrator user can quickly make a decision based on the recommended list of products for advertisement placement, and accurately. Products with high sales potential and good results that can be achieved after the products are advertised are selected for advertising.

It can be known from the typical embodiments of the present application that the technical solutions of the present application have many advantages, including but not limited to the following aspects:

First, the present application calculates the corresponding recommendation score by obtaining the adoption rate and effectiveness data of the predicted product, and then constructs a recommended list of advertisement placement products according to the recommendation score and pushes it to the management user of the independent site. The difference from the prior art is only Relying on the performance data of the predicted products as the reference index for advertising product selection, the technical solution here integrates the product’s corresponding adoption rate and the performance data of the two dimensions of the recommendation score obtained as the reference index for advertising product selection, which can accurately predict the product selection. The probability of placing advertisements, and predicting the probability of conversions that may be harvested after the products are advertised, reflect the degree of suitability of the products for advertisements more comprehensively with quantified values, which is convenient for the management of independent sites. Users can quickly make decisions based on the recommendation score, and select the best choice. Products that are suitable for advertising and can achieve good results after advertising are selected as advertising options.

Secondly, on the one hand, most independent sites have limited financial resources and are unable to have most of the rich product advertisement data, resulting in the inability to decide suitable products for advertisement. The advertisement recommendation list can effectively assist this type of advertisement The management user of the independent site selects the products for advertisement placement on the products of the independent site, so as to solve the real pain point of the management user; The recommendation list can more effectively assist such novice users to select suitable products on their own independent sites, so as to place advertisements on them, improve user experience, and increase user stickiness.

In addition, the technical solution of the present application is simple and easy to implement, the time complexity of model calculation is low, and the engineering can be parallelized on a large scale.

Referring to FIG. 2, in a further embodiment, step S1100, before the step of acquiring the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, includes the following steps:

Step S1000, determining the adoption rate of each commodity in the commodity database of the independent site by adopting a pre-trained adoption rate prediction model to a convergent state;

Invoke the corresponding data access interface of the commodity database of the independent site provided by the e-commerce service platform, obtain the commodity information corresponding to each commodity from the commodity database, and input it into the adoption rate prediction model pre-trained to the convergence state, so as to be determined by The model determines the adoption rate for each item.

The commodity information includes but is not limited to commodity pictures, commodity titles, commodity details, commodity prices, commodity sku (category), inventory, and commodity shelf time. The adoption rate prediction model is pre-trained to a convergent state so that it has the ability to predict the adoption rate corresponding to each commodity based on the commodity information of the commodity. Specifically, the model can extract the commodity of the commodity. In the information, the management user of the independent site is urged to select the product characteristics and statistical characteristics corresponding to the advertisement of the product, and determine the adoption rate. Commodities and commodity prices are low, and the statistical characteristics include but are not limited to high click-through rate, high repurchase rate, high sales volume, and high purchase rate after the commodity is advertised.

Invoke the pre-trained to convergent adoption rate prediction model to perform feature extraction on each commodity in the commodity database of the independent site, extract the commodity characteristics and the statistical characteristics, thereby obtaining the output two-way feature vector, The two-way feature vectors are respectively the product feature vector and the statistical feature vector. Further, the product feature vector and the feature vector are fused to obtain a fused feature vector, the fused feature vector is mapped to the classification space, and then the classifier constructed by Softmax or Sigmod is used to pair the Probability calculation is performed on the classification space to obtain the probability that the fusion feature vector is mapped to the classification space corresponding to the product selected for advertisement, that is, the adoption rate.

The process from training to convergence of the procurement rate prediction model is further disclosed in the embodiments in the subsequent parts of this application, and this step is not listed below for the time being.

Step S1010, determining the click-through rate of each commodity in the commodity database of the independent site by using the click-through rate prediction model pre-trained to a convergent state;

Invoke the corresponding data access interface of the commodity database of the independent site provided by the e-commerce service platform, obtain the commodity information corresponding to each commodity from the commodity database, and input it into the click-through rate prediction model pre-trained to the convergence state, so as to be determined by The model determines the click-through rate for each item.

The commodity information includes but is not limited to commodity pictures, commodity titles, commodity details, commodity prices, commodity sku (category), inventory, and commodity shelf time. The click-through rate prediction model is pre-trained to a convergent state so that it has the ability to prompt users to click on the corresponding product features in the product information of the product after the product is advertised to determine the click-through rate of the product, and the product features include image features and text. feature.

Invoke the pre-trained to convergent click-through rate prediction model to perform feature extraction on the commodity information corresponding to each commodity in the commodity database of the independent site, and extract the commodity information of the commodity to urge the user to click the commodity to place the advertisement corresponding to the commodity. The product features of the model are obtained, the feature vector output by the model is obtained, the feature vector is mapped to the classification space, and then the classifier constructed by Softmax or Sigmod is used to perform probability calculation on the classification space, and the feature vector is obtained. The probability of the category space being clicked by the user is the click-through rate.

The process from training to convergence of the click-through rate prediction model is further disclosed in the following embodiments of this application, and this step is not listed below for the time being.

Step S1020, determining the conversion rate of each commodity in the commodity database of the independent site by using the conversion rate prediction model trained in advance to the convergence state;

Invoke the corresponding data access interface of the commodity database of the independent site provided by the e-commerce service platform, obtain commodity information corresponding to each commodity from the commodity database, and input it into the conversion rate prediction model pre-trained to the convergence state, so as to be determined by The model determines the conversion rate for each item.

The commodity information includes but is not limited to commodity pictures, commodity titles, commodity details, commodity prices, commodity sku (category), inventory, and commodity shelf time.

The conversion rate prediction model is pre-trained to a convergent state so that it has the ability to generate product characteristics corresponding to other user behaviors and determine the conversion rate of the product according to the product information of the product after the product is placed on the advertisement and clicked by the user. The product features include image features and text features.

Invoke the pre-trained to convergent conversion rate prediction model to perform feature extraction on the commodity information corresponding to each commodity in the commodity database of the independent site, and extract the commodity information of the commodity after the advertisement of the commodity is clicked by the user. , to promote the production of product features corresponding to other user behaviors, obtain the feature vector output by the model, map the feature vector to the classification space, and then use the classifier constructed by Softmax or Sigmod to perform probability calculation on the classification space to obtain the feature vector. The conversion rate is the probability mapped to the classification space of other user behaviors after the advertisement is clicked by the user.

The process from training to convergence of the conversion rate prediction model is further disclosed in the embodiments in the subsequent parts of this application, and this step is not listed below for the time being.

Step S1030 , store the adoption rate and effectiveness data of each commodity in the commodity database corresponding to each commodity, where the effectiveness data is the product of the click-through rate and the conversion rate associated with the same commodity.

The performance data includes the click-through rate and conversion rate corresponding to the product.

Both the adoption rate and the effectiveness data of each product are time-sensitive. Therefore, a timed task can be set, and the timed task is set to execute steps S1000-S1020 at a specific time point. For example, , the specific time point can be in the early morning of every day, because the commodity data in the commodity database at this time changes less, and the resources that can be allocated by the server of the e-commerce service platform are sufficient, so it is very suitable for executing the scheduled task, Accordingly, in order to keep the adoption rate and performance data properly updated to make them more accurate, it lays a foundation for the accuracy of ad placement selection in subsequent steps.

In order to facilitate the subsequent invocation of the adoption rate and effectiveness data of each commodity, the commodities corresponding to their mapping associations are stored in the database. For the effectiveness data, the click-through rate and the effectiveness data associated with the same commodity are stored in the database. The product of the conversion rate is expressed as a numerical value of the performance data.

In this embodiment, the purchase data and performance data corresponding to each product in the product database of the independent site are prepared in advance, so that the subsequent command or request for calling the purchase data and performance data can be quickly responded to, which greatly improves the response efficiency and reduces the user's waiting time. , to improve user experience and increase user stickiness.

Referring to FIG. 3, in a further embodiment, in step S1000, in the step of determining the adoption rate of each commodity in the commodity database of the independent site by using the adoption rate prediction model pre-trained to a convergent state, the training of the adoption rate prediction model process, including the following steps:

Step S1001, calling a training sample from the data set, the training sample includes a positive sample and a negative sample, each training sample is provided with a supervision label indicating whether the product is advertised, and the positive sample includes the product database of the e-commerce platform. The product information and statistical information of the advertised products in the e-commerce platform, the negative sample includes the product information and statistical information of the products that have not been advertised in the product database of the e-commerce platform, and the product information includes the product shelf time, product category, product Price, the statistical information includes click-through rate, purchase rate, and repurchase rate;

In one embodiment, a text preprocessing operation is performed on the training samples in the data set in order to facilitate the adoption rate prediction model to more accurately extract relevant features from the training samples in the data set, and the text preprocessing operation includes removing the training samples. For example, use regular expressions to delete the non-text part of the text, and delete some special non-Chinese and English characters and punctuation marks; perform word segmentation on the training samples. For English, the word segmentation can directly call the split() function, that is Yes, for Chinese, the stuttering word segmentation method is adopted, and the corresponding method can be called directly by the third-party platform. The pre-packaged method function data package can directly call the method in it; remove the stop words in the training sample, and remove the Stop words have no effect on the overall semantics, such as a large number of function words, pronouns or nouns and verbs with no specific meaning in Chinese, but articles and pronouns in English.

The ratio of positive and negative samples is flexibly set by those skilled in the art based on prior knowledge or experimental experience, and it is recommended to be 1:10, wherein the statistical information in the positive samples includes the corresponding advertised products. The click rate, purchase rate, and repurchase rate of , the statistical information in the negative sample is empty.

Step S1002, inputting the training sample into the adoption rate prediction model to obtain a commodity feature vector and a statistical feature vector corresponding to the commodity information and statistical information in the training sample;

Input the training sample into the adoption rate prediction model for feature extraction, and extract the product features and statistics corresponding to the product information and statistical information of the positive and negative sample products to prompt the management user of the independent site to select the product for advertisement. feature, obtain the product feature vector and statistical feature vector output by the model, the product features include but are not limited to the products with a short time on the shelf, the products of the hot-selling category in the current product sales market, and the low prices of the products, and the statistical features include but are not limited to Limited to high click rate, high repurchase rate, high sales volume and high purchase rate after the product is advertised, one hot coding is used for each feature in the corresponding product features and statistical features, and the coding in Word2vec and Bert can also be used here. module, so as to obtain two-way embedding codes corresponding to the output, that is, the commodity feature vector and the statistical feature vector.

Step S1003, splicing the commodity feature vector and the statistical feature vector to obtain a feature fusion vector, and mapping its input fully connected layer to a preset classification space to obtain a prediction adoption rate;

The splicing is vector splicing, and feature fusion is performed on the commodity feature vector and the statistical feature vector. Generally speaking, in order to fuse the two feature vectors, the two are usually directly related to the corresponding elements in the same dimension. Add and obtain the feature fusion vector. If the dimensions of the two are different, they can be converted into vectors of the same dimension through linear transformation. Further, the feature fusion vector is input into the fully connected layer for linear transformation, and it is mapped to two The classification space, the binary space includes the positive class space where the representative product is selected for advertising, and the negative class space where the representative product is not selected for advertising, and then the classifier is used to calculate the probability that the feature fusion vector is mapped to the second class space, This probability is obtained as the predicted acceptance rate by a Sigmod or Softmax method function in this process.

Step S1004: Calculate the loss value corresponding to the cross-entropy loss of the predicted adoption rate according to the supervision label, determine whether the loss value reaches a preset threshold, and terminate the training when it reaches the preset threshold; otherwise, according to the loss The gradient update is performed on the model, and the next training sample in the data set is called to continue the iterative training of the model.

Call the preset loss function, which can be set flexibly by those skilled in the art based on prior knowledge or experimental experience, and calculate the loss value corresponding to the cross-entropy loss of the predicted adoption rate based on the supervision label. When the loss value reaches When the preset threshold is used, it indicates that the model has been trained to a convergent state, so that the model training can be terminated; when the loss value does not reach the preset threshold, it indicates that the model has not converged, so the model is updated according to the loss value. Gradient update, usually backpropagation correction weight parameters of each link of the model to make the model further approach convergence, and then continue to call the next sample data in the data set to perform iterative training on the model until the model is trained to a convergent state.

Those skilled in the art should know that the adoption rate prediction model is supervised and trained based on the data set composed of the supervised labels and positive and negative samples until the model converges, so that the adoption rate prediction model has the commodity information according to the commodity The ability to predict the corresponding adoption rate of each product, specifically, the model can prompt the management user of the independent site to select the product feature corresponding to the product advertisement according to the product information extracted from the product database. and statistical features to obtain the corresponding adoption rate of the product. The product features include, but are not limited to, the product has a short shelf life, the current product sales market is a hot-selling product, and the product price is low. The statistical feature includes the corresponding product after the product is advertised. High click rate, high repurchase rate, high sales volume and high purchase rate.

In this embodiment, the training process of the adoption rate prediction model based on the deep learning model is disclosed. It can be seen that under the training of the sample data of the data set and the supervised label, the adoption rate prediction model extracts the product information corresponding to the product. By inferring the commodity characteristics and statistical characteristics of the commodity, it can obtain a more accurate prediction ability, so as to ensure that it can serve the prediction of the corresponding adoption rate of the commodity in the future.

Referring to FIG. 4, in a further embodiment, in step S1010, in the step of determining the click-through rate of each commodity in the commodity database of the independent site using a click-through rate prediction model pre-trained to a convergent state, the training of the click-through rate prediction model The process includes the following iterative steps:

Step S1011: Acquire training samples in the data set, where the training samples include commodity information of commodities that have been advertised and clicked by users in the commodity database of the e-commerce platform, and the corresponding supervision indicating whether the advertisements of the commodities are clicked by users. Label;

The commodity information includes but is not limited to commodity pictures, commodity titles, commodity details, commodity prices, commodity sku (category), inventory, and commodity shelf time. In one embodiment, a text preprocessing operation is performed on the training samples in the data set in order to facilitate the neural network model to more accurately extract relevant features from the training samples in the data set, and the text preprocessing operation includes removing the data in the training samples. For example, use regular expressions to delete some special non-Chinese and English characters and punctuation marks; perform word segmentation on training samples. For English, the word segmentation can directly call the split() function. , for Chinese, the stammering word segmentation method is adopted, and the corresponding method can be called directly by the third-party platform. The pre-packaged method function data package can directly call the method in it; remove the stop words in the training sample, and remove the stop word. The use of words has no effect on the overall semantics, such as a large number of function words, pronouns or nouns and verbs without specific meanings in Chinese, while articles and pronouns in English.

Step S1012, inputting the training sample into the click-through rate prediction model to extract corresponding deep semantic features to obtain feature vectors;

Inputting the training samples into the click-through rate prediction model for feature extraction, extracting the product features in the product information of the product that urge the user to click the corresponding product features after the product is advertised, and extracting the image features corresponding to the product pictures in the product information. In the process of extracting features, the product image is first divided into multiple image units, and the size of each image unit is equal. Therefore, after the feature image feature extraction is performed by the click-through rate prediction model, each image unit is A corresponding single-image feature vector can be obtained, and then, the output of the click-through rate prediction model is converted into a high-dimensional feature vector, and the image is spliced according to the corresponding position information of each high-dimensional feature vector corresponding to the image element in the product image. Encoding vector; feature extraction is performed on the preprocessed text information such as the product title and product details in the product information, and the corresponding text features are extracted. One hot encoding is used for each feature. The encoding can also use Word2vec, The encoding module in Bert, thereby obtaining the embedding code corresponding to the model output, that is, the commodity text encoding vector. Further, the image encoding vector and the commodity text encoding vector are vector spliced to perform feature fusion. Generally speaking, in order to fuse the The two encoding vectors are usually directly added with corresponding elements in the same dimension to obtain a fusion encoding vector. If the dimensions of the two are different, they can be converted into vectors of the same dimension through linear transformation.

The click-through rate prediction model structure includes a picture feature extraction module and a text feature extraction module, and the two modules can be derived from a neural network model based on CNN that is suitable for deep semantic feature extraction of pictures, such as Resnet, EfficientNet Corresponding feature extraction modules in other models, and neural network models based on RNN that are suitable for deep semantic feature extraction of text, such as the corresponding feature extraction modules in Bert, Electra and other models.

Step S1013, using a classifier to classify and map the feature vector, and predict the corresponding click-through rate;

Further, the fusion encoding vector is input to the fully connected layer for linear transformation, and it is mapped to a binary space, the binary space includes a positive space where advertisements representing products are clicked by users, and a negative space where ads representing products are clicked. If the advertisement is not clicked by the user, the classifier is used to calculate the probability that the fusion coding vector is mapped to the second-class space. In this process, the probability is obtained through the Sigmod or Softmax method function as the predicted click-through rate.

Step S1014: Calculate the cross-entropy loss value corresponding to the click-through rate predicted by the classifier according to the supervision label, and perform gradient update on the model according to the cross-entropy loss value until the model converges.

Call the preset loss function, which can be set flexibly by those skilled in the art according to prior knowledge or experimental experience, and calculate the loss value corresponding to the cross-entropy loss of the predicted click rate based on the supervision label. When the loss value reaches When the preset threshold is used, it indicates that the model has been trained to a convergent state, so that the model training can be terminated; when the loss value does not reach the preset threshold, it indicates that the model has not converged, so the model is updated according to the loss value. Gradient update, usually backpropagation correction weight parameters of each link of the model to make the model further approach convergence, and then continue to call the next sample data in the data set to perform iterative training on the model until the model is trained to a convergent state.

It can be understood that the CTR prediction model is supervised and trained based on the supervision label until the model converges, so that the model can extract the product features corresponding to the product information of the product that prompts the product to be advertised and clicked by the user, and determine the product. The corresponding click rate, the product features include text features and image features.

In this embodiment, the training process of the click-through rate prediction model based on the deep learning model is disclosed. It can be seen that under the training of the sample data of the data set and the supervision label, the click-through rate prediction model extracts the product information of the product. Inference from the image features and text features can obtain more accurate prediction ability, so as to ensure that it can serve the prediction of the corresponding click rate of the product in the future, so that the click rate can more accurately reflect that the product is clicked by the user after the advertisement is placed. possibility.

Referring to FIG. 5, in a further embodiment, in step S1020, in the step of determining the conversion rate of each commodity in the commodity database of the independent site by using a conversion rate prediction model pre-trained to a convergent state, the training of the conversion rate prediction model The process includes the following iterative steps:

Step S1021: Obtain training samples in the data set, where the training samples include commodity information of commodities that have been advertised in the commodity database of the e-commerce platform and have been clicked by the user to generate relevant user behaviors, and the advertisements representing the commodities that have been clicked by the user Whether the supervision label corresponding to the relevant user behavior is generated afterward;

The commodity information includes but is not limited to commodity pictures, commodity titles, commodity details, commodity prices, commodity sku (category), inventory, and commodity shelf time. The relevant user behaviors include but are not limited to adding to the shopping cart, purchasing, and repurchasing.

In one embodiment, a text preprocessing operation is performed on the training samples in the data set in order to facilitate the neural network model to more accurately extract relevant features from the training samples in the data set, and the text preprocessing operation includes removing the data in the training samples. For example, use regular expressions to delete some special non-Chinese and English characters and punctuation marks; perform word segmentation on training samples. For English, the word segmentation can directly call the split() function. , for Chinese, the stammering word segmentation method is adopted, and the corresponding method can be called directly by the third-party platform. The pre-packaged method function data package can directly call the method in it; remove the stop words in the training sample, and remove the stop word. The use of words has no effect on the overall semantics, such as a large number of function words, pronouns or nouns and verbs without specific meanings in Chinese, while articles and pronouns in English.

Step S1022, inputting the training sample into the conversion rate prediction model to extract corresponding deep semantic features to obtain feature vectors;

Input the training sample into the conversion rate prediction model for feature extraction, and extract the product features corresponding to other user behaviors after the product advertisement is clicked by the user in the product information of the product. In the feature extraction process of the image features corresponding to the product images, the product image is firstly divided into a plurality of image units, and the size of each image unit is equal. Therefore, after the feature image feature extraction is performed by the conversion rate prediction model , each image unit can obtain a corresponding single image feature vector, and then convert the output of the conversion rate prediction model into a high-dimensional feature vector. The location information is spliced into a picture encoding vector; feature extraction is performed on the preprocessed text information such as the product title and product details in the product information, and the corresponding text features are extracted. Onehot encoding is used for each feature, which is encoded here. The encoding module in Word2vec and Bert can also be used to obtain the embedding encoding corresponding to the model output, that is, the commodity text encoding vector. Further, the image encoding vector and the commodity text encoding vector are vector spliced to perform feature fusion. Generally, In other words, in order to fuse the two coding vectors, the fused coding vector is obtained by directly adding the corresponding elements of the two in the same dimension. If the dimensions of the two are different, they can be converted into the same dimension by linear transformation. vector.

The conversion rate prediction model structure includes a picture feature extraction module and a text feature extraction module, which can be derived from a CNN-based neural network model suitable for deep semantic feature extraction of pictures, such as Resnet, EfficientNet Corresponding feature extraction modules in other models, and neural network models based on RNN that are suitable for deep semantic feature extraction of text, such as the corresponding feature extraction modules in Bert, Electra and other models.

Step S1023, using a classifier to classify and map the feature vector to predict the corresponding conversion rate;

Further, the fusion coding vector is input to the fully connected layer for linear transformation, and it is mapped to a binary space. The space that represents the advertisement of the product is clicked by the user and does not produce relevant user behavior, and then the classifier is used to calculate the probability that the fusion coding vector is mapped to the second-class space. In this process, the probability is obtained through the Sigmod or Softmax method function as the Predicted conversion rate.

Step S1024: Calculate the cross-entropy loss value corresponding to the conversion rate predicted by the classifier according to the supervision label, and perform gradient update on the model according to the cross-entropy loss value until the model converges.

Call the preset loss function, which can be set flexibly by those skilled in the art according to prior knowledge or experimental experience, and calculate the loss value corresponding to the cross-entropy loss of the predicted conversion rate based on the supervision label. When the loss value reaches When the preset threshold is used, it indicates that the model has been trained to a convergent state, so that the model training can be terminated; when the loss value does not reach the preset threshold, it indicates that the model has not converged, so the model is updated according to the loss value. Gradient update, usually backpropagation correction weight parameters of each link of the model to make the model further approach convergence, and then continue to call the next sample data in the data set to perform iterative training on the model until the model is trained to a convergent state.

It can be understood that the conversion rate prediction model is supervised and trained based on the supervision label until the model converges, so that the model can extract the product information in the product information. The product features of the product are determined to determine the corresponding conversion rate of the product, and the product features include text features and image features.

In this embodiment, the training process of the conversion rate prediction model based on the deep learning model is disclosed. It can be seen that under the training of the sample data of the data set and the supervision labels, the conversion rate prediction model extracts the commodity information of the commodity. Inference from the image features and text features of the product can obtain a more accurate prediction ability, so as to ensure that the subsequent prediction of the corresponding conversion rate of the product can be used, so as to more accurately reflect the conversion rate of the product after the advertisement is clicked by the user. Possibility to generate relevant user behavior.

Referring to FIG. 6 , in an extended embodiment, step S1300 , selecting a commodity in the commodity database as an advertisement placement selection according to the recommendation score to construct an advertisement placement selection recommendation list, including the following steps:

Step S1310, sorting according to the recommendation scores corresponding to the commodities in the commodity database, and constructing a list of recommended products for advertisement placement;

According to the recommendation scores corresponding to the commodities in the commodity database, a plurality of commodities with the highest corresponding recommendation scores are selected as advertisement placement items, and the commodities are sorted in descending order of the recommendation scores corresponding to the advertisement placement options, Then, an advertisement placement recommendation list is constructed based on the advertisement placement selections in this order, and the number of selected commodities can be flexibly changed by those skilled in the art according to the number of commodities displayed in the advertisement placement selection recommendation list.

In an extended embodiment, three sorts of other three dimensions can be set for the advertisement placement recommendation list, and the three sortings can be switched with the aforementioned recommendation score sorting, and the advertisement placement selection recommendation can be set. The list is displayed in the order of the aforementioned recommendation scores by default, and it is optional to switch between the three kinds of ordering. Price, the corresponding established products are sorted in order from early to late; the first letter of the product category is sorted in the order of A-Z, and the same category is classified together; the price of the product is sorted from low to high, so that the The three sorts are the optional switching display of the recommended list of advertisement placement options. In this way, it is convenient for the user to quickly select suitable products from the recommended product selection list for advertisement placement based on the ranking of recommendation scores and the characteristics corresponding to the other three rankings. For example, the user can select the recommended score ranking from the recommended product selection list for advertisement placement. Third, the commodity price ranks first, that is, the commodity with the lowest price and high estimated sales potential.

Step S1320: In response to an advertisement recommendation request from an online store, push the advertisement placement recommendation list to the management user of the independent site.

The independent site runs its own store, that is, an online store. The management user of the independent site can touch the advertisement recommendation button set up by the e-commerce service platform to provide advertisement recommendation services on the online store. The advertisement recommendation request is sent to the server of the e-commerce service platform, and the server receives the request and responds to it, and pushes the advertisement placement recommendation list to the management user of the independent site.

In this embodiment, according to the recommendation scores corresponding to the commodities in the commodity database, a list of recommended products for advertisement placement is constructed and pushed to the management user of the independent site, so that the management user can make quick decisions and make accurate selections. Produce products with certain sales potential and better advertising results.

Please refer to FIG. 7 , a device for selecting products for advertisement placement provided for one of the purposes of the present application is a functional embodiment of the method for selecting products for placement of advertisements in the present application. The device includes: a data acquisition module 1100 and a score calculation module 1200 , and the list building module 1300, wherein the data acquisition module 1100 is used to obtain the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the advertisement prediction data includes the adoption rate and effect data, and the adoption rate represents the The predicted probability of the product being advertised, and the performance data includes the conversion probability representing the possible harvest after the product is advertised; the score calculation module 1200 is used to calculate the corresponding product corresponding to each product according to the adoption rate of the product and the performance data. The list building module 1300 is configured to select commodities in the commodity database as advertisement placement options according to the recommendation scores to construct a recommended list of advertisement placement options.

In a further embodiment, before the data acquisition module 1100, it includes: an adoption rate determination module for determining the adoption rate of each commodity in the commodity database of the independent site by using the adoption rate prediction model pre-trained to a convergent state; click-through rate; The confirmation module is used to determine the click-through rate of each commodity in the commodity database of the independent site by using the click-through rate prediction model pre-trained to the convergence state; the conversion rate confirmation module is used to use the conversion rate prediction model pre-trained to the convergence state to determine the independent The conversion rate of each commodity in the commodity database of the site; the data storage sub-module is used to store the adoption rate of each commodity and the effectiveness data corresponding to each commodity in the commodity database, and the effectiveness data is related to the same commodity. The product of the click-through rate and the conversion rate.

In a further embodiment, in the adoption rate determination module, the training process of the adoption rate prediction model includes: a training sample invoking module for invoking a training sample from the data set, and the training sample includes a positive sample and a Negative samples, each training sample is provided with a supervision label indicating whether the product is advertised, the positive sample includes the product information and statistical information of the advertised products in the product database of the e-commerce platform, and the negative sample includes the electronic product. The commodity information and statistical information of the unadvertised commodities in the commodity database of the merchant platform, the commodity information includes the commodity shelf time, commodity category, commodity price, and the statistical information includes the click rate, the purchase rate, and the repurchase rate; feature extraction module for inputting the training sample into the adoption rate prediction model to obtain commodity feature vectors and statistical feature vectors corresponding to the commodity information and statistical information in the training sample; a vector splicing module for splicing the commodity features vector and the statistical feature vector to obtain a feature fusion vector, and map its input fully connected layer to a preset classification space to obtain a prediction adoption rate; an iterative training module is used to calculate the crossover of the prediction adoption rate according to the supervision label. The loss value corresponding to the entropy loss, determine whether the loss value reaches the preset threshold, and when it reaches the preset threshold, the training is terminated; otherwise, the gradient update is performed on the model according to the loss value, and the next training in the data set is called. The samples continue to perform iterative training on the model.

In a further embodiment, in the click-through rate confirmation module, the iterative training process of the click-through rate prediction model includes: a training sample calling module for acquiring training samples in the data set, and the training samples include The commodity information of the commodity that has been advertised and clicked by the user in the commodity database, and the corresponding supervision label indicating whether the advertisement of the commodity has been clicked by the user; the feature extraction module is used to input the training sample into the click-through rate prediction. The model extracts corresponding deep semantic features to obtain feature vectors; the classification mapping module is used to classify and map the feature vectors by using a classifier to predict the corresponding click rate; the gradient update module is used to calculate according to the supervision label The cross-entropy loss value corresponding to the click-through rate predicted by the classifier is obtained, and the gradient update is performed on the model according to the cross-entropy loss value until the model converges.

In a further embodiment, in the conversion rate confirmation module, the iterative training process of the conversion rate prediction model includes: a training sample invoking module for acquiring training samples in the data set, and the training samples include Commodity information of commodities that have been advertised and clicked by users to generate relevant user behaviors in the commodity database, as well as the supervision labels that indicate whether the advertisements of the commodity have been clicked by users to generate relevant user behaviors; the feature extraction module is used to extract The training sample is input into the conversion rate prediction model to extract corresponding deep semantic features, and a feature vector is obtained; a classification mapping module is used to classify and map the feature vector by using a classifier, so as to predict the corresponding conversion rate; gradient; An update module, configured to calculate a cross-entropy loss value corresponding to the conversion rate predicted by the classifier according to the supervision label, and perform gradient update on the model according to the cross-entropy loss value until the model converges.

In an extended embodiment, the list building module 1300 includes: a score sorting module, used for sorting according to the recommendation scores corresponding to the commodities in the commodity database, and constructing a list of recommended products for advertisement placement; a list push module, used for In response to an advertisement recommendation request from an online store, the advertisement placement recommendation list is pushed to the management user of the independent site.

To solve the above technical problems, the embodiments of the present application also provide computer equipment. As shown in FIG. 8 , a schematic diagram of the internal structure of the computer equipment. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected by a system bus. Wherein, 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, and when the computer-readable instructions are executed by the processor, the processor can be made to implement a Advertising selection method. The processor of the computer equipment is used to provide computing and control capabilities and support the operation of the entire computer equipment. Computer-readable instructions may be stored in the memory of the computer device, and when the computer-readable instructions are executed by the processor, the processor may execute the method for selecting advertisements of the present application. The network interface of the computer equipment is used for communication with the terminal connection. Those skilled in the art can understand that the structure shown in FIG. 8 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. Include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.

In this embodiment, the processor is used to execute the specific functions of each module and its sub-modules in FIG. 7 , and the memory stores 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. The memory in this embodiment stores the program codes and data required to execute all modules/sub-modules in the advertisement placement device of the present application, and the server can call the server's program codes and data to execute the functions of all sub-modules.

The present application further 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 can execute the method for selecting advertisements in any embodiment of the present application. A step of.

The present application also provides a computer program product, including computer programs/instructions, when the computer program/instructions are executed by one or more processors, to implement the steps of the method described in any embodiment of the present application.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above-mentioned embodiments of the present application can be implemented by instructing 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 may include the flow of the embodiments of the above-mentioned methods. The aforementioned storage medium may be a computer-readable storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM).

Those skilled in the art can understand that various operations, methods, steps, measures, and solutions in the process discussed in this application may be alternated, modified, combined or deleted. Further, other steps, measures, and solutions in the various operations, methods, and processes that have been discussed in this application may also be alternated, modified, rearranged, decomposed, combined, or deleted. Further, steps, measures and solutions in the prior art with various operations, methods, and processes disclosed in this application may also be alternated, modified, rearranged, decomposed, combined or deleted.

To sum up, this application accurately predicts the probability of a product being selected for advertisement, the probability of being clicked by a user after a product placement advertisement is exposed, and the product placement advertisement being clicked by a user through a purchase rate prediction model, a click-through rate prediction model, and a conversion rate prediction model. The probability of corresponding user behaviors is then generated, and these probabilities accurately reflect the market competitiveness of the product, that is, the sales potential, and the results that can be achieved after the product is advertised. The advertisement placement list quickly makes decisions on the selection of advertisement placement products, and accurately selects products with certain sales potential and better advertisement placement results.

The above are only part of the embodiments of the present application. It should be pointed out that for those skilled in the art, without departing from the principles of the present application, several improvements and modifications can also be made. It should be regarded as the protection scope of this application.

Claims (10)

1. a method for selecting products for advertisement placement, is characterized in that, comprises the steps: Obtain the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the advertisement prediction data includes the adoption rate and the effectiveness data, the adoption rate represents the predicted probability of the commodity being advertised, and the effectiveness data includes the characteristics representing the commodity The conversion probability that may be harvested after being placed in the advertisement; According to the adoption rate and effectiveness data of the product, the recommendation score corresponding to each product is calculated; According to the recommendation score, commodities in the commodity database are selected as advertisement placement options to construct an advertisement placement option recommendation list. 2. The method according to claim 1, characterized in that, before the step of acquiring the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the method comprises the following steps: Determine the adoption rate of each commodity in the commodity database of the independent site using an adoption rate prediction model pre-trained to a convergent state; Determine the click-through rate of each item in the item database of the independent site using a CTR prediction model pre-trained to a convergent state; Determine the conversion rate of each item in the item database of the independent site using a conversion rate prediction model pre-trained to a convergent state; The adoption rate and effectiveness data of each commodity are stored in the commodity database corresponding to each commodity, and the effectiveness data is the product of the click-through rate and the conversion rate associated with the same commodity. 3. The method for selecting products according to claim 2 is characterized in that, in the step of determining the adoption rate of each commodity in the commodity database of the independent site by adopting the adoption rate prediction model trained in advance to the convergence state, the adoption rate is The training process of the rate prediction model includes the following steps: A training sample is called from the data set. The training sample includes positive samples and negative samples. Each training sample is provided with a supervision label indicating whether the product is advertised. The product information and statistical information of the advertised products, the negative sample includes the product information and statistical information of the unadvertised products in the product database of the e-commerce platform, and the product information includes the product shelf time, product category, and product price. The above statistical information includes click rate, purchase rate and repurchase rate; Inputting the training sample into the adoption rate prediction model to obtain a commodity feature vector and a statistical feature vector corresponding to the commodity information and statistical information in the training sample; Splicing the commodity feature vector and the statistical feature vector to obtain a feature fusion vector, and mapping its input fully connected layer to a preset classification space to obtain a prediction adoption rate; Calculate the loss value corresponding to the cross-entropy loss of the predicted adoption rate according to the supervision label, and judge whether the loss value reaches the preset threshold. When it reaches the preset threshold, the training is terminated; otherwise, the training is terminated according to the loss value. The model implements gradient update, and the next training sample in the data set is called to continue the iterative training of the model. 4. The method for selecting products according to claim 2, characterized in that, in the step of determining the click-through rate of each commodity in the commodity database of an independent site using a click-through rate prediction model trained in advance to a convergent state, the click The training process of the rate prediction model includes the following iterative steps: Acquiring training samples in the data set, where the training samples include commodity information of commodities that have been advertised and clicked by the user in the commodity database of the e-commerce platform, and a supervision label indicating whether the advertisement of the commodity has been clicked by the user; Inputting the training sample into the click-through rate prediction model to extract corresponding deep semantic features to obtain feature vectors; Use a classifier to classify and map the feature vector, and predict the corresponding click-through rate; A cross-entropy loss value corresponding to the click-through rate predicted by the classifier is calculated according to the supervision label, and a gradient update is performed on the model according to the cross-entropy loss value until the model converges. 5. The method for selecting products for advertisement placement according to claim 2, characterized in that, in the step of determining the conversion rate of each commodity in the commodity database of an independent site using a conversion rate prediction model trained in advance to a convergent state, the conversion The training process of the rate prediction model includes the following iterative steps: Obtain training samples in the data set, where the training samples include commodity information of commodities that have been advertised in the commodity database of the e-commerce platform and have been clicked by users to generate relevant user behaviors, and whether the advertisements for the commodities are generated after being clicked by users Supervision labels corresponding to relevant user behaviors; Inputting the training sample into the conversion rate prediction model to extract corresponding deep semantic features to obtain feature vectors; Use a classifier to classify and map the feature vector to predict the corresponding conversion rate; A cross-entropy loss value corresponding to the conversion rate predicted by the classifier is calculated according to the supervision label, and a gradient update is performed on the model according to the cross-entropy loss value until the model converges. 6. The method for selecting products for advertisement placement according to claim 1, characterized in that, selecting commodities in the commodity database as advertisement placement products selection according to the recommendation score to construct an advertisement placement product selection recommendation list, comprising the following steps: Sort according to the recommendation scores corresponding to the commodities in the commodity database, and construct a recommendation list for advertisement placement; In response to an advertisement recommendation request from an online store, the advertisement placement recommendation list is pushed to the management user of the independent site. 7. A text translation device, comprising: The data acquisition module is used to acquire the advertisement prediction data corresponding to each commodity in the commodity database of the independent site, the advertisement prediction data includes the adoption rate and the effect data, the adoption rate represents the prediction probability of the commodity being advertised, and the The performance data includes the conversion probability that characterizes the possible harvest after the product is advertised; A score calculation module, configured to calculate the recommendation score corresponding to each product according to the adoption rate and effectiveness data of the product; The list building module is configured to select the commodities in the commodity database as advertisement placement options according to the recommendation score to construct a recommended list of advertisement placement options. 8. A computer device comprising 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 computer program according to any one of claims 1 to 6 steps of the method. 9. A computer-readable storage medium, characterized in that it stores a computer program implemented by the method according to any one of claims 1 to 6 in the form of computer-readable instructions, and the computer program is called by a computer At runtime, the steps included in the corresponding method are executed. 10. A computer program product comprising a computer program/instruction, characterized in that, when the computer program/instruction is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images