CN116186412A

CN116186412A - Recommendation information display method and device

Info

Publication number: CN116186412A
Application number: CN202310228572.4A
Authority: CN
Inventors: 胡素芸; 宋弋芳
Original assignee: Shanghai Bilibili Technology Co Ltd
Current assignee: Shanghai Bilibili Technology Co Ltd
Priority date: 2023-03-08
Filing date: 2023-03-08
Publication date: 2023-05-30

Abstract

The application provides a recommendation information display method and device, wherein the recommendation information display method comprises the following steps: obtaining at least one candidate recommendation information, estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information, determining the target style in the first style dimension according to the conversion rate, generating a target display template of the first candidate recommendation information according to the target style in each style dimension of the first candidate recommendation information, sorting the candidate recommendation information based on the target display templates of the candidate recommendation information, and displaying the recommendation information according to the sorting result. Through carrying out fine division on the pattern dimension, the conversion rate of the candidate patterns in each pattern dimension is estimated, and the target patterns with high conversion rate are determined according to the estimated conversion rate, so that the target display template with high conversion rate is generated, the accuracy of the sequencing result is improved, more accurate recommended content is provided for a user, the display cost is reduced, and the user experience is improved.

Description

Recommendation information display method and device

Technical Field

The application relates to the technical field of recommended information, in particular to a recommended information display method. The application also relates to a recommended information presentation apparatus, a computing device, and a computer-readable storage medium.

Background

With the development of internet technology, more and more recommended information is widely applied to platforms such as applications, applets, webpages and the like, and recommended content meeting the user requirements is provided for users.

At present, because the candidate recommendation information is numerous, the candidate recommendation information needs to be reasonably ordered and displayed, so that more accurate recommendation content is provided for users.

However, the ranking is directly performed according to the conversion rate of the recommendation information, the conversion rate of the recommendation information is ignored to be influenced by various factors, the ranking result is inaccurate, the displayed recommendation information is difficult to provide accurate recommendation content for users, and the user experience is insufficient.

Disclosure of Invention

In view of this, the embodiment of the application provides a recommendation information display method. The application relates to a recommended information display device, a computing device and a computer readable storage medium, so as to solve the problems of inaccurate recommended content and insufficient user experience caused by inaccurate sequencing results of recommended information in the prior art.

According to a first aspect of an embodiment of the present application, there is provided a recommendation information display method, including:

acquiring at least one candidate recommendation information, wherein the candidate recommendation information is composed of at least one style dimension, and each style dimension comprises at least one candidate style;

Estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information, and determining a target style in the first style dimension according to the conversion rate, wherein the first candidate recommendation information is any one of at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information;

generating a target display template of the first candidate recommendation information according to the target style of each style dimension of the first candidate recommendation information;

and sorting the candidate recommendation information based on the target display templates of the candidate recommendation information, and displaying the recommendation information according to the sorting result.

According to a second aspect of embodiments of the present application, there is provided a recommended information presentation apparatus, including:

an acquisition module configured to acquire at least one candidate recommendation information, wherein the candidate recommendation information is composed of at least one style dimension, and each style dimension comprises at least one candidate style;

the first determining module is configured to estimate conversion rates of the candidate styles in a first style dimension of the first candidate recommendation information, and determine a target style in the first style dimension according to the conversion rates, wherein the first candidate recommendation information is any one of at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information;

The second determining module is configured to generate a target display template of the first candidate recommendation information according to the target styles of the first candidate recommendation information in each style dimension;

and the display module is configured to sort the candidate recommendation information based on the target display templates of the candidate recommendation information and display the recommendation information according to the sorting result.

According to a third aspect of embodiments of the present application, there is provided a computing device comprising a memory, a processor and computer instructions stored on the memory and executable on the processor, the processor implementing the steps of the recommended information presentation method when executing the instructions.

According to a fourth aspect of embodiments of the present application, there is provided a computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the recommendation information presentation method.

The recommendation information display method provided by the application,

In one or more embodiments of the present application, at least one candidate recommendation information is obtained, where the candidate recommendation information is composed of at least one style dimension, at least one candidate style is included in each style dimension, a conversion rate of each candidate style in a first style dimension of the first candidate recommendation information is estimated, a target style in the first style dimension is determined according to the conversion rate, the first candidate recommendation information is any one of the at least one candidate recommendation information, the first style dimension is any style dimension of the first candidate recommendation information, a target display template of the first candidate recommendation information is generated according to the target style in each style dimension of the first candidate recommendation information, each candidate recommendation information is ordered based on the target display templates of each candidate recommendation information, and recommendation information display is performed according to an ordering result. Finely dividing according to pattern dimensions forming candidate recommendation information, estimating conversion rate corresponding to at least one candidate pattern in each pattern dimension, determining a target pattern with high conversion rate in the pattern dimension based on the conversion rate, generating a target display template with high conversion rate according to the target pattern with high conversion rate in each pattern dimension, and more accurately sequencing and displaying each candidate recommendation information based on the target pattern template with high conversion rate, so that more accurate recommendation content is provided for users, and user experience is improved.

Drawings

FIG. 1 is a flowchart of a method for displaying recommendation information according to an embodiment of the present application;

FIG. 2 is a schematic diagram illustrating conversion rate correction in a recommended information presentation method according to an embodiment of the present disclosure;

FIG. 3 is a system flow chart of a method for displaying recommendation information according to an embodiment of the present application;

FIG. 4 is a schematic flow chart of determining a target display template in a recommendation information display method according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of a conversion rate estimation model in a recommended information display method according to an embodiment of the present application;

fig. 6 is a schematic diagram of determining a target style in a recommendation information display method according to an embodiment of the present application;

FIG. 7 is a process flow diagram of a recommendation information display method applied to platform advertisements according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of a propulsion information display device according to an embodiment of the present disclosure;

fig. 9 is a block diagram of a computing device according to an embodiment of the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is, however, susceptible of embodiment in many other ways than those herein described and similar generalizations can be made by those skilled in the art without departing from the spirit of the application and the application is therefore not limited to the specific embodiments disclosed below.

The terminology used in one or more embodiments of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of one or more embodiments of the application. As used in this application in one or more embodiments and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present application refers to and encompasses any or all possible combinations of one or more of the associated listed items.

It should be understood that, although the terms first, second, etc. may be used in one or more embodiments of the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, a first may also be referred to as a second, and similarly, a second may also be referred to as a first, without departing from the scope of one or more embodiments of the present application. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.

First, terms related to one or more embodiments of the present application will be explained.

Recommendation information: the information with the recommending function is used for front-end display, the recommending information is provided with a corresponding display template, and the display template is provided with recommending contents in different styles. Including but not limited to search results, goods, and advertisements. For example, the user inputs index words on a certain search engine webpage, and displays a plurality of search results arranged according to a specific order to the user on a landing page, for example, the user inputs goods index words on a certain e-commerce platform applet, and displays a plurality of goods arranged according to a specific order to the user on the landing page, for example, after entering a certain video platform application, the user displays a plurality of advertisements arranged according to a specific order to the user on a first page.

Information flow advertisement: friend dynamics located in social media applications (web pages), or advertisements in content streams of information media applications (web pages, applets) and audiovisual media applications (web pages). The information flow advertisement display template comprises different types of pictures, texts, videos and the like, different information flow advertisements are marked through a specific algorithm, the information flow advertisements are directionally put according to marked labels, and the conversion rate of the information flow advertisements is improved, wherein key influence factors of the conversion rate comprise the display template, target objects and resource input, and the subsequent embodiments of the information flow advertisement display template are also based on the ranking of one or more key influence factors.

Recommendation information engine system: and the online information recommendation system is used for directionally recalling initial recommendation information from the recommendation information base, obtaining a sequencing result of the recommendation information through sequencing, and displaying the recommendation information according to the sequencing result.

Coarse row: and screening the initial recommendation information of the directional recall in the engine system, aiming at meeting the requirements of hardware performance and cost, rapidly screening the initial recommendation information to obtain candidate recommendation information, and performing subsequent fine-ranking processing.

Fine discharge: in the engine system, the recommendation information is used for determining the accurate ranking result of the candidate recommendation information aiming at meeting the suitability with the user, and displaying the recommendation information according to the ranking result.

CTR (Click Through Rate): and browsing the conversion rate, and representing the probability of clicking by the user after the conversion rate is displayed on the page to the user. In the case where the recommendation information is a search result, a good, or an advertisement, CTR is a click conversion rate, that is, a probability of clicking correspondingly.

CVR (ClickValue Rate): and operating the conversion rate, and representing the probability of selecting the corresponding recommended content by the user after the page is displayed to the user. When the recommendation information is an advertisement, the CVR is an advertisement conversion rate, that is, a probability of performing a behavior such as purchasing, collecting, or adding a shopping cart, and the like, and corresponds to a purchase conversion rate, a collection conversion rate, or a shopping cart addition conversion rate.

CTCVR (Click Through ClickValue Rate): and operating the browsing conversion rate, and representing the probability of operating the recommended content of the recommended information, specifically the multiplied result of CTR and CVR, after the user browses.

Thompson sampling: the method for sampling probability density according to Beta distribution is applied to Thompson sampling of the conversion rate of recommended information, and comprises the following specific modes: determining alpha parameters of Beta distribution as the clicking times of a user after the recommended information is displayed on a page, and determining Beta parameters of Beta distribution as the non-clicking times of the user after the recommended information is displayed on the page; acquiring alpha parameters and beta parameters of each candidate pattern; determining a random number for each of the alpha parameter and the Beta parameter using Beta distribution; sorting according to the random numbers, and outputting candidate patterns corresponding to the maximum values; according to the user behavior, alpha parameters and beta parameters are adjusted (the user clicks, the alpha parameters are added with 1, the user does not click, and the beta parameters are added with 1); and according to the finally adjusted alpha parameter and beta parameter, randomly extracting by using a probability density function, and determining the conversion rate of the candidate pattern.

DPA (Dynamic Product Ads, dynamic cargo advertisement): and adaptively adjusting the goods advertisement according to the user portrait to generate the dynamic advertisement.

LR (Logistic Regression ) model: the linear regression model consists of a linear regression model and a sigmoid function, and the output result of the linear regression model is a continuous value.

DNN (Deep Neural Networks, deep neural network) model: the input layer, the hidden layer and the output layer are included, the input layer, the hidden layer and the output layer are provided with a plurality of hidden layers, and all the hidden layers are connected through a full connection mode.

FFM (Field-aware Factorization Machines, field perceptron factorizer) model: each feature of the FFM corresponds to not a unique hidden vector, but a set of hidden vectors. And when the feature cross is performed, performing inner product operation on each feature and the hidden vector corresponding to the opposite domain to obtain the weight of the cross feature.

In the field of recommendation information, a display template is an important feature for finely arranging candidate recommendation information, and the display template of the recommendation information comprises patterns such as large images, small images, text links, videos and the like. The recommendation information of the video display template is a special recommendation information, and has a specific style, for example, a specific object video is displayed on the video display template. Other component styles, such as text connections, are presented in the play page or comment area. Thus, the different styles in the presentation template are also an important feature. For a plurality of style dimensions of the video recommendation information, for example, display delay time, button style, text style, component overall style and the like, after determining a target style dimension from each style dimension, generating a display template determined by the target style dimension as a target display template of the video recommendation information, wherein different display templates have different display effects and correspond to different conversion rates. How to determine a display template highly adapted to the recommended information, so as to obtain a high conversion rate of the recommended information, further obtain a more accurate sequencing result and display the result, and ensure that more accurate recommended content is provided for users is a problem to be solved.

At present, one method is to autonomously select target patterns in different types of dimensions by a provider of recommended information, generate corresponding target display templates according to the target patterns in each type of dimensions, and the other method is to provide selectable candidate patterns in different types of dimensions by a display platform of the recommended information, determine rules according to preset patterns, match the target patterns for the recommended information, and generate corresponding target display templates. However, in the former method, the provider needs to spend a lot of time in the process of autonomously selecting the target pattern, and in order to obtain the display effect of high conversion rate, data analysis needs to be performed on different candidate patterns, which increases the delivery cost, and in addition, increases the operation cost when performing iterative updating of the display template. In another method, due to the number of users and the number of recommended information, it is difficult for the preset style determination rule to determine a target presentation template of high conversion rate for each recommended information in consideration of its own specificity.

In view of the foregoing, in the present application, a recommended information display method is provided, and the present application relates to a recommended information display device, a computing apparatus, and a computer-readable storage medium, which are described in detail in the following embodiments one by one.

Fig. 1 shows a flowchart of a recommendation information display method according to an embodiment of the present application, which specifically includes the following steps:

step 102: at least one candidate recommendation information is obtained, wherein the candidate recommendation information is composed of at least one style dimension, and each style dimension comprises at least one candidate style.

The embodiment of the application is applied to the server of the application, the applet or the web page with the recommendation information ordering function, and a recommendation information engine system is deployed on the server.

The candidate recommendation information is recommendation information provided by a provider stored in a recommendation information base, and the candidate recommendation information can be obtained directly from the recommendation information base or obtained by pre-screening after being obtained from the recommendation information base, and is not limited herein. Candidate recommendation information has specific information content including, but not limited to: the method comprises the steps of relevant information of recommended content corresponding to candidate recommendation information, the type of a display template of the candidate recommendation information and provider information of the candidate recommendation information. It should be noted that, the types of the display templates of the candidate recommendation information may be the same or different, for example, the type of the display template of the candidate recommendation information 1 is video, and the type of the display template of the candidate recommendation information 2 is graphics context. As another example, the creative type of the presentation template of the candidate recommendation information is associated with the presentation carrier of the recommendation information, including a programmed creative type, a DPA creative type, and a landing page creative type.

The style dimension is a type dimension of each style constituting the candidate recommended information, for example, different types dimensions of styles such as presentation delay time, button style, text style, component whole style, and the like. Note that, the style dimensions of each candidate recommendation information may be the same or different, for example, style dimensions of candidate recommendation information 1 include: the style dimensions of the candidate recommendation information 2 include: picture resolution and text style.

The candidate patterns are a plurality of patterns selectable in a pattern dimension, for example, when the pattern dimension is a presentation delay time, the candidate patterns in the pattern dimension are: 15ms,30ms,60ms.

Illustratively, 100 candidate recommendation information rcd_info is obtained from the recommendation information base database_rcd_info: (rcd_info_i, i e [1,100 ]), wherein the candidate recommendation information rcd_info_i is composed of n Style dimensions style_info_list: (Style_Info_list_j, j e [1, n ]) comprising m candidate styles Style_Info under any Style dimension Style_Info_list_j: (Style_Info_k, k.epsilon.1, m).

At least one candidate recommendation information is obtained, wherein the candidate recommendation information is composed of at least one pattern dimension, each pattern dimension comprises at least one candidate pattern, and an information foundation, a dimension foundation and a pattern foundation are laid for subsequently estimating the conversion rate of each candidate pattern and determining a target pattern in each pattern dimension according to the conversion rate.

Step 104: estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information, and determining the target style in the first style dimension according to the conversion rate, wherein the first candidate recommendation information is any one of at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information.

In step 104, the conversion rate is used as the determination basis of the target pattern, so that the determined target pattern is the preferred conversion rate pattern in each pattern dimension, and is a unified measurement standard determined for different candidate recommendation information, different pattern dimensions and different candidate patterns.

The conversion rate may be any one of the conversion rates such as a browsing conversion rate, an operation conversion rate, and an operation browsing conversion rate of the candidate style, or may be a conversion rate obtained by further classifying and weighting calculation based on the conversion rates, for example, a target conversion rate may be obtained by weighting calculation of the browsing conversion rate, the operation conversion rate, and the operation browsing conversion rate based on preset weights (W1, W2, and W3).

The conversion rate of each candidate style in the first style dimension of the first candidate recommendation information is estimated, specifically, the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information is estimated according to the user information. Since the conversion rate varies significantly from user to user, for example, the user is a young user whose conversion rate for a larger font style in the search results is higher for the elderly than for the young. Therefore, it is necessary to estimate the conversion rate of each candidate style based on the user information. The method for estimating the conversion rate may be to estimate statistics of the historical conversion rate according to the estimation, for example, linear programming statistics, or estimate by using a neural network model, which is not limited herein.

And determining the target patterns in the first pattern dimension according to the conversion rate, wherein the method comprises the steps of sorting all candidate patterns in the first pattern dimension according to the conversion rate, and determining the target patterns in the first pattern dimension according to the sorting result. Further, the target pattern in the first pattern dimension may be determined according to a preset threshold, where the preset threshold may be a number threshold, a conversion rate threshold, or a directly determined first-order candidate pattern may be the target pattern.

Illustratively, ctctcvrs of the respective candidate styles style_info_k under the first type dimension style_info_list_1 of the first candidate recommendation information rcd_info_1 are estimated according to statistics of the history CTCVRs, the respective candidate styles style_info_k under the first type dimension style_info_list_1 are ordered according to CTCVRs of the respective candidate styles style_info_k, and the candidate styles style_info_4 of the first order is determined as Target Style style_style_info under the first type dimension style_info_list_1 according to the ordering result.

Estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information, and determining the target style in the first style dimension according to the conversion rate, wherein the first candidate recommendation information is any one of at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information. The conversion rate of the candidate patterns in each pattern dimension which is finely divided is estimated, and the target patterns with high conversion rate in each pattern dimension are obtained according to the estimated conversion rate, so that a foundation is laid for the subsequent generation of the target display template with high conversion rate.

Step 106: and generating a target display template of the first candidate recommendation information according to the target style of each style dimension of the first candidate recommendation information.

The target display template is a pattern combination of target patterns under each pattern dimension of candidate recommendation information, and the display template is determined. For example, the target styles under each style dimension (video play button style, video subtitle style, and video play speed style) of the candidate recommendation information 1 are: a style 1 style play button, a bold and bold subtitle style, 5 optional play speeds (0.25 x,0.5x,1x,1.5x,2 x), and a target presentation template (video play button style: style 1 style; video subtitle style: bold, video play speed style: 5 play speeds (0.25 x,0.5x,1x,1.5x,2 x)) is determined based on the 3 target styles.

Generating a target display template of the first candidate recommendation information according to the target styles of the first candidate recommendation information in each style dimension, specifically, generating the target display template of the first candidate recommendation information according to the style combination of the target styles of the first candidate recommendation information in each style dimension.

Illustratively, the Target presentation template target_pattern_rcd_info_1 of the first candidate recommendation information rcd_info_1 is generated according to Pattern combinations of Target patterns (style_info_list_1: style_info_4; style_info_list_2: style_info_2 … … style_info_list_n: style_info_7) in n Pattern dimensions of the first candidate recommendation information rcd_info_1.

And generating a target display template of the first candidate recommendation information according to the target style of each style dimension of the first candidate recommendation information. And further generating a target display template with high conversion rate according to the target style with high conversion rate in each style dimension, and providing a basis for subsequent sorting.

Step 108: and sorting the candidate recommendation information based on the target display templates of the candidate recommendation information, and displaying the recommendation information according to the sorting result.

The recommendation information is displayed as front-end display of the recommendation information, and the display carrier corresponds to the type of the display template of the candidate recommendation information.

And ordering the candidate recommendation information based on the target display templates of the candidate recommendation information, in a specific mode, ordering the candidate recommendation information based on the target display templates and the information content of the candidate recommendation information.

The recommendation information display can be performed according to the sorting result, the target recommendation information can be obtained by screening according to the sorting result, the target recommendation information can be displayed, the candidate recommendation information can be displayed sequentially according to the sorting result, the sequential display can be up-and-down sliding or page turning full display in the same page, or can be independent display in a card form, and the switching is performed through sliding.

Illustratively, based on the Target display templates of 100 candidate recommendation information rcd_info, target_pattern_rcd_info and information Content rcd_info_content, the 100 candidate recommendation information rcd_info is ranked, and page-turning comprehensive display is performed on the 100 candidate recommendation information rcd_info according to the ranking result.

In the embodiment of the application, at least one candidate recommendation information is obtained, wherein the candidate recommendation information is composed of at least one style dimension, at least one candidate style is included in each style dimension, the conversion rate of each candidate style in a first style dimension of the first candidate recommendation information is estimated, the target style in the first style dimension is determined according to the conversion rate, the first candidate recommendation information is any one of the at least one candidate recommendation information, the first style dimension is any style dimension of the first candidate recommendation information, a target display template of the first candidate recommendation information is generated according to the target style in each style dimension of the first candidate recommendation information, each candidate recommendation information is ordered based on the target display template of each candidate recommendation information, and recommendation information display is performed according to an ordering result. The method comprises the steps of finely dividing according to pattern dimensions forming candidate recommendation information, estimating conversion rate corresponding to at least one candidate pattern in each pattern dimension, determining a target pattern with high conversion rate in the pattern dimension according to the conversion rate, generating a target display template with high conversion rate according to the target pattern with high conversion rate in each pattern dimension, and more accurately sequencing and displaying each candidate recommendation information based on the target pattern template with high conversion rate, so that more accurate recommendation content is provided for a user, display cost is reduced, and user experience is improved.

Optionally, estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information in step 104 includes the following specific steps:

and estimating the conversion rate of each candidate pattern in the first pattern dimension of the first candidate recommendation information by utilizing a pre-trained conversion rate estimation model according to the user information, wherein the conversion rate estimation model is a neural network model obtained by training according to the sample user information, the sample pattern and the historical conversion rate of the sample pattern.

The user information is identity information of the user corresponding to the recommendation information, can be user information of the user corresponding to each candidate recommendation information, and can also be user information of a user union of each candidate recommendation information.

The conversion rate estimation model is a neural network model with feature extraction and conversion rate calculation functions, and is obtained by training according to sample user information, sample patterns and historical conversion rates of the sample patterns in advance. For example, LR models, DNN models, FFN models, and the like.

Sample user information, sample patterns, and historical conversion rates for sample patterns are determined from historical behavioral data. For example, the history behavior data is obtained by browsing, operating or browsing operation processing on recommended information for 10000 users, wherein the recommended information is formed by a plurality of sample patterns, the user information of 10000 users is determined as sample user information, and the history conversion rate is determined according to the behavior data.

The specific pre-training method is as follows:

obtaining a pre-training sample set, wherein the pre-training sample set comprises a plurality of sample groups, and each sample group comprises sample user information, sample patterns and historical conversion rates of the sample patterns; extracting a first sample group from the pre-training sample set, wherein the first sample group is any one of a plurality of sample groups, the first sample group comprising first sample user information, a first sample pattern, and a historical conversion rate of the first sample pattern; according to the first sample user information, predicting the predicted conversion rate of the first sample pattern by using a conversion rate estimation model; calculating a loss value according to the predicted conversion rate and the historical conversion rate; and (3) according to the loss value, adjusting model parameters of the conversion rate estimation model, and returning to execute the extraction of the first sample group from the pre-training sample set, and obtaining the conversion rate estimation model after training under the condition that the preset loss value threshold value is reached.

According to user information, the conversion rate of each candidate pattern in the first pattern dimension of the first candidate recommendation information is estimated by using a pre-trained conversion rate estimation model.

It should be noted that, because training defects (such as a sample, a loss value determination, a parameter adjustment method, etc.) may exist in the pre-training, there is a limitation on the model performance of the conversion rate estimation model, and when the estimated conversion rate is lower than the preset judgment threshold value, thompson sampling is performed to correct the estimated conversion rate. The steps of the thompson sampling are as follows: determining alpha parameters of Beta distribution as the clicking times of a user after the recommended information is displayed on a page, and determining Beta parameters of Beta distribution as the non-clicking times of the user after the recommended information is displayed on the page; acquiring alpha parameters and beta parameters of each candidate pattern; determining a random number for each of the alpha parameter and the Beta parameter using Beta distribution; sorting according to the random numbers, and outputting candidate patterns corresponding to the maximum values; according to the user behavior, alpha parameters and beta parameters are adjusted (the user clicks, the alpha parameters are added with 1, the user does not click, and the beta parameters are added with 1); and according to the finally adjusted alpha parameter and beta parameter, randomly extracting by using a probability density function, and determining the conversion rate of the candidate pattern.

Fig. 2 is a schematic diagram illustrating conversion rate correction in a recommended information presentation method according to an embodiment of the present application.

As shown in fig. 2, at different alpha and beta values: probability density function curve (1): α=β=0.5; probability density function curve (2): α=5, β=1; probability density function curve (3): α=1, β=3; probability density function curve (4): α=2, β=2; probability density function curve (5): α=2, β=5, corresponding to 5 probability density function curves.

Illustratively, using a pre-trained DNN model, ctctctvr of each candidate Style info_k is estimated based on Style characteristics feature_style_info_k of each candidate Style info_k under a first Style dimension style_info_list_1 of User information user_info and first candidate recommendation information rcd_info_1.

And estimating the conversion rate of each candidate pattern in the first pattern dimension of the first candidate recommendation information by utilizing a pre-trained conversion rate estimation model according to the user information, wherein the conversion rate estimation model is a neural network model obtained by training according to the sample user information, the sample pattern and the historical conversion rate of the sample pattern. Based on user information, the conversion rate of each candidate pattern is predicted by using a pre-trained conversion rate estimation model, so that the suitability between the conversion rate and the user information is improved, the accuracy of obtaining the conversion rate by prediction is improved, the accuracy of determining the target pattern subsequently is ensured, and the processing efficiency is improved.

Optionally, the conversion rate estimation model comprises a user feature extraction module, a style feature extraction module and a calculation module;

correspondingly, according to the user information, the conversion rate of each candidate pattern under the first pattern dimension of the first candidate recommendation information is estimated by utilizing a pre-trained conversion rate estimation model, and the method comprises the following specific steps of:

extracting the characteristics of the user information by using a user characteristic extraction module to obtain user characteristics;

extracting features of each candidate style in the first style dimension of the first candidate recommendation information by using a style feature extraction module to obtain style features of each candidate style;

and according to the user characteristics and each style characteristic, calculating the conversion rate of each candidate style by using a calculation module.

The conversion rate estimation model is a neural network model with a double-tower structure and comprises a user characteristic extraction module, a style characteristic extraction module and a calculation module, wherein each module is obtained by training corresponding sample data in advance. The user characteristic extraction module and the style characteristic extraction module are two mutually independent sub-network modules, and the two modules can independently extract the obtained user characteristics and style characteristics and independently cache the user characteristics and style characteristics. The calculation module is a functional module for carrying out corresponding feature calculation according to the user features and the style features to obtain the corresponding conversion rate.

Illustratively, a User Feature extraction module of the DNN model is utilized to perform Feature extraction on User information user_info to obtain User Feature feature_user_info, a pattern Feature extraction module of the DNN model is utilized to perform Feature extraction on each candidate pattern style_info_k under a first pattern dimension style_info_list_1 of the first candidate recommendation information rcd_info_1 to obtain pattern Feature feature_style_info_k of each candidate pattern style_info_k, and a calculation module is utilized to perform Feature inner product calculation according to the User Feature feature_user_info and each pattern Feature feature_style_info_k to obtain conversion rate of each candidate pattern style_info_k.

Extracting the characteristics of the user information by using a user characteristic extraction module to obtain user characteristics; extracting features of each candidate style in the first style dimension of the first candidate recommendation information by using a style feature extraction module to obtain style features of each candidate style; and according to the user characteristics and each style characteristic, calculating the conversion rate of each candidate style by using a calculation module. Further improving the estimated efficiency of the conversion rate and the accuracy of the conversion rate, and ensuring the follow-up determination of a more accurate target pattern with high conversion rate.

Optionally, determining the target style in the first style dimension according to the conversion rate in step 104 includes the following specific steps:

according to the information content of the user information and at least one candidate recommendation information, carrying out validity screening on each candidate style in the first style dimension;

and determining the target style in the first style dimension according to the screening result and the conversion rate.

Before determining the target style, the effectiveness test needs to be performed on each candidate style, for example, the candidate recommendation information is an image-text type advertisement of a certain cargo, the text style is a bold text with a large word size, and if the target display template is determined according to the text style, the image of the cargo is covered.

The screening result is a candidate pattern that satisfies the validity screening condition.

And carrying out validity screening on each candidate pattern in the first pattern dimension according to the information content of the user information and the at least one candidate recommendation information. The preset validity screening condition is a preset judging condition for determining user information and information content and judging whether the display rule of the candidate pattern is met or not. For example, the information content includes text information of candidate recommended information and information of recommended content, and for text advertisement of a certain cargo, the text information and cargo information of the cargo determine whether the display rule of the candidate pattern is satisfied according to the user information, the text information and the cargo information.

And determining the target pattern in the first pattern dimension according to the screening result and the conversion rate, wherein the screening result is ordered according to the conversion rate, and the target pattern in the first pattern dimension is determined according to the ordering result. The method for determining the target pattern from the screening result is discussed in detail in step 104, and will not be described here.

Illustratively, according to the User information user_info and the information Content info_content of the candidate recommendation information, validity screening is performed on each candidate Style style_info_k under the first Style dimension style_info_list_1 by using preset validity screening conditions, screening results are ordered according to conversion rates, and the Target Style target_style_info under the first Style dimension style_info_list_1 is determined according to the ordering results.

According to the information content of the user information and at least one candidate recommendation information, carrying out validity screening on each candidate style in the first style dimension; and determining the target style in the first style dimension according to the screening result and the conversion rate. The effectiveness of the target patterns is improved, the effectiveness of the target display templates generated later is further improved, and more accurate sequencing and display are realized.

Optionally, in step 108, sorting the candidate recommendation information based on the target display templates of the candidate recommendation information includes the following specific steps:

based on target display templates and information contents of each candidate recommendation information, pre-trained recommendation information estimation models are utilized to estimate information conversion rates of the candidate recommendation information under each target display template, wherein the recommendation information estimation models are neural network models obtained through training according to the display templates and the information contents of the sample recommendation information and the historical conversion rates of the sample recommendation information;

and ordering each candidate recommendation information according to the information conversion rate.

The recommended information estimation model is a neural network model with feature extraction and information conversion rate calculation functions, and is obtained by training according to a display template and information content of sample recommended information and the historical conversion rate of the sample recommended information in advance. For example, LR models, DNN models, FFN models, and the like. It should be noted that the recommended information estimation model may be a double-tower structure model having a user information feature extraction function and a recommended information feature extraction function, and the recommended information estimation model is obtained by performing adaptive pre-training on the fine-pitch model. The information features of the display template are added on the basis of the fine-ranking model, and the estimation of the information conversion rate is participated.

The specific pre-training method is as follows:

the method comprises the steps of obtaining a pre-training sample set, wherein the pre-training sample set comprises a plurality of sample groups, and each sample group comprises a display template and information content of sample recommendation information and a historical conversion rate of the sample recommendation information; extracting a first sample group from the pre-training sample set, wherein the first sample group is any one of a plurality of sample groups, and comprises a display template of first sample recommendation information, information content and a historical conversion rate; the display template and the information content of the first sample recommendation information are used for estimating the prediction information conversion rate of the first sample recommendation information by using a recommendation information estimation model; calculating a loss value according to the predicted information conversion rate and the historical conversion rate; and (3) according to the loss value, adjusting model parameters of the recommended information estimation model, returning to execute the extraction of the first sample group from the pre-training sample set, and obtaining the trained recommended information estimation model under the condition that the preset loss value threshold value is reached.

The display template and the information content of the sample recommendation information and the historical conversion rate of the sample recommendation information are determined according to the historical behavior data. For example, the historical behavior data is that 1000000 users browse, operate or browse operation processing is performed on the recommended information, wherein 1000000 recommended information is determined as sample recommended information.

Based on the target display templates and the information content of each candidate recommendation information, the information conversion rate of the candidate recommendation information under each target display template is estimated by utilizing a pre-trained recommendation information estimation model.

Illustratively, using the pre-trained FFM model, based on the information features feature_target_pattern_rcd_info and the information features feature_rcd_info_content of the Target presentation templates target_pattern_rcd_info of 100 candidate recommendation information rcd_info, the information conversion rates of candidate recommendation information rcd_info under the 100 Target presentation templates target_pattern_rcd_info are estimated, and the respective candidate recommendation information rcd_info_i is ordered according to the information conversion rates.

Based on target display templates and information contents of each candidate recommendation information, pre-trained recommendation information estimation models are utilized to estimate information conversion rates of the candidate recommendation information under each target display template, wherein the recommendation information estimation models are neural network models obtained through training according to the display templates and the information contents of the sample recommendation information and the historical conversion rates of the sample recommendation information; and ordering each candidate recommendation information according to the information conversion rate. The accuracy of the sequencing result is further improved, and more accurate recommended content is provided for the user.

Optionally, sorting the candidate recommendation information according to the information conversion rate includes the following specific steps:

screening each candidate recommendation information according to a preset information online strategy to obtain screened candidate recommendation information;

and sorting the screened candidate recommendation information according to the information conversion rate.

The preset information online strategy is an application of the recommended information ordering function, an applet or a recommended information online screening strategy of a network page. For example, if a recommendation information has a user age limit and cannot be displayed in priority, the recommendation information is filtered.

Screening each candidate recommendation information according to a preset information online strategy and the information content of each candidate recommendation information to obtain screened candidate recommendation information

Illustratively, according to a preset information online strategy (prohibited display of tobacco and wine cargoes) and cargo information of 100 candidate recommendation information rcd_info, screening each candidate recommendation information rcd_info_i to obtain screened candidate recommendation information rcd_info, and sorting the screened candidate recommendation information rcd_info according to the information conversion rate.

Screening each candidate recommendation information according to a preset information online strategy to obtain screened candidate recommendation information; and sorting the screened candidate recommendation information according to the information conversion rate. The rules of on-line preset information are met, the display effect is improved, and the user experience is improved.

Optionally, in step 108, the presenting of the recommended information according to the sorting result includes the following specific steps:

determining target recommendation information from at least one candidate recommendation information according to the sorting result;

and displaying the target recommendation information.

According to the sorting result, the target recommendation information is determined from at least one candidate recommendation information, specifically, the target recommendation information can be determined according to a preset threshold, the preset threshold can be a number threshold, can be an information conversion rate threshold, and can be directly determined as the target recommendation information.

Illustratively, the candidate recommendation information rcd_info_57 of the first cis-position is determined as the Target recommendation information target_rcd_info according to the sorting result.

Determining target recommendation information from at least one candidate recommendation information according to the sorting result; the target recommendation information is displayed, the target recommendation information with high conversion rate is determined to be displayed, more accurate recommendation content is provided for the user, and user experience is improved.

Optionally, step 102 includes the following specific steps:

acquiring at least one candidate recommendation information, including:

acquiring a plurality of initial recommendation information;

and screening and obtaining at least one candidate recommendation information from the plurality of initial recommendation information according to a preset information screening strategy.

The initial recommendation information is recalled recommendation information directed from a recommendation information base of the recommendation information engine system.

The preset information screening policy is a judgment policy for pre-screening recalled recommended information, is preset according to hardware performance and cost, and can be preset screening judgment conditions, a coarse row model, or a combination of the screening judgment conditions and the coarse row model, which is not limited herein. The coarse row model may be an LR model, a DNN model, and an FFM model, among others.

Illustratively, 10000 initial recommendation information rcd_info' are recalled from the recommendation information base database_rcd_info orientation: (rcd_info_i ', i e [1,10000 ]), screening judgment conditions (the recommended information size is not more than 500 MB) and an LR model, and screening 100 candidate recommended information rcd_info from 10000 initial recommended information rcd_info'. (rcd_info_i, i.e. [1,100 ]).

Acquiring a plurality of initial recommendation information; and screening and obtaining at least one candidate recommendation information from the plurality of initial recommendation information according to a preset information screening strategy. The data volume of the conversion rate estimation performed later is reduced, and the display efficiency is improved.

Fig. 3 is a system flowchart of a recommendation information display method according to an embodiment of the present application.

As shown in fig. 3, initial recommendation information is obtained through directional recall, and then coarse ranking is performed on the initial recommendation information after screening, so as to obtain candidate recommendation information, wherein the candidate recommendation information comprises different types of display templates (a programmed display template, a DPA display template and a landing page display template), a target display template is determined through steps 104-106, fine ranking is performed on the candidate recommendation information under the target display template, target recommendation information is determined, and resource investment is performed on the target recommendation information.

Fig. 4 is a schematic flow chart of determining a target display template in a recommendation information display method according to an embodiment of the present application.

As shown in fig. 4, the recommendation information queue is obtained through coarse-rank filtering, and before participating in fine-rank of the model, for each style dimension of the component style of the presentation template, for example, delay, shopping cart style, text color, and sub-card type … …. Creating a candidate pattern queue in each pattern dimension from the candidate patterns for each presentation template participating in fine ranking (including a programmed presentation template, a DPA presentation template and the like), for example, creating a delay pattern queue for the delay pattern dimension, and creating n candidate pattern queues for the n pattern dimensions, wherein each candidate pattern queue contains candidate pattern sets of all presentation templates requested this time, for example, for the delay pattern queue including delay pattern set 1, delay pattern set 2 and delay pattern set 3 … …, m candidates are filled with m candidate patterns style_info, a single candidate Style style_info represents a component pattern of the presentation template in the pattern dimension, for example, for delay pattern set 3 including delay pattern 1, delay pattern 2 and decoction pattern 3 … …, and for each candidate pattern, data including a pre-estimated method, pattern mark, alpha and beta values generated by an offline strategy for calculating a pre-estimated forest sampling value, presentation template id and the like is included.

Based on different style dimensions, the recommendation information is ranked by using a double-tower model, so that a ranking result of each style dimension, for example, a delay ranking result, a shopping cart style ranking result, a text color ranking result and a daughter card type ranking result is obtained, and an optimal solution of each style dimension is obtained.

Fig. 5 is a schematic structural diagram of a conversion rate estimation model in a recommendation information display method according to an embodiment of the present application.

As shown in fig. 5, the user information is subjected to feature extraction by using the user feature extraction module to obtain user features, the recommended information and the candidate patterns are subjected to feature extraction by using the pattern feature extraction module to obtain pattern features, and the operation browsing conversion rate is calculated based on the user features and the pattern features.

Fig. 6 is a schematic diagram illustrating determination of a target style in a recommendation information presentation method according to an embodiment of the application.

As shown in fig. 6, for any candidate style set of the first candidate styles, which includes a plurality of candidate styles, the candidate styles are ranked according to the conversion rate (conversion rate 1, conversion rate 2, conversion rate 3 … …), so as to obtain a ranking result as follows: candidate pattern 1, candidate pattern 2 and candidate pattern 3, carrying out validity test on each candidate pattern, determining that the candidate pattern meeting the validity test is candidate pattern 2 and candidate pattern 3 … …, and if the candidate pattern not meeting the validity test has candidate pattern 1, determining that the candidate pattern 2 with the highest conversion rate is the target pattern in the pattern dimension.

The application of the recommendation information display method provided in the present application to platform advertisement is taken as an example, and the recommendation information display method is further described below with reference to fig. 7. Fig. 7 shows a process flow chart of a recommendation information display method applied to platform advertisement according to an embodiment of the present application, which specifically includes the following steps:

step 702: a plurality of initial platform advertisements are recalled directionally from an advertisement library of the advertisement engine system;

step 704: screening at least one candidate platform advertisement from a plurality of initial platform advertisements according to preset advertisement filtering conditions and a coarse-ranking model;

step 706: extracting the characteristics of the user information by using a user characteristic extraction module of the pre-trained DNN model to obtain the user characteristics;

step 708: carrying out feature extraction on each candidate style under a plurality of style dimensions of the candidate platform advertisement by using a style feature extraction module of the pre-trained DNN model to obtain style features of each candidate style;

step 710: according to the user characteristics and the style characteristics, calculating the characteristic inner product of the user characteristics and the style characteristics by utilizing a calculation module of a pre-training DNN model to obtain the conversion rate of each candidate style;

Step 712: determining target patterns in a plurality of pattern dimensions according to the conversion rate;

step 714: generating a target display template of at least one candidate platform advertisement according to the plurality of target patterns;

step 716: based on a target display template and information content of the candidate platform advertisement, estimating the information conversion rate of the candidate platform advertisement under the target display template by utilizing a pre-trained fine-ranking model;

step 718: sequencing at least one candidate platform advertisement according to a preset information online strategy and an information conversion rate;

step 720: determining a target platform advertisement from at least one candidate platform advertisement according to the sorting result;

step 722: and performing front-end display on the target platform advertisement.

According to the embodiment of the application, the conversion rate of the candidate patterns under the multi-dimensions is refined and estimated, the target patterns under the multi-dimensions with higher conversion rate are determined, the target display templates with higher conversion rate are generated according to the target patterns, the time cost, the throwing cost and the operation cost of a provider of platform advertisements are greatly saved, the fine-ranking model is utilized, the target platform advertisements under the target display templates are optimized, front-end display is carried out, more accurate recommended content is provided for users, and user experience is improved.

Corresponding to the method embodiment, the present application further provides a recommended information display device embodiment, and fig. 8 shows a schematic structural diagram of a recommended information display device according to an embodiment of the present application. As shown in fig. 8, the apparatus includes:

an obtaining module 802 configured to obtain at least one candidate recommendation information, wherein the candidate recommendation information is composed of at least one style dimension, and each style dimension includes at least one candidate style;

a first determining module 804, configured to estimate a conversion rate of each candidate style in a first style dimension of the first candidate recommendation information, and determine a target style in the first style dimension according to the conversion rate, where the first candidate recommendation information is any one of the at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information;

a second determining module 806 configured to generate a target presentation template of the first candidate recommendation information according to the target styles in the respective style dimensions of the first candidate recommendation information;

and the display module 808 is configured to sort the candidate recommendation information based on the target display templates of the candidate recommendation information, and display the recommendation information according to the sorting result.

Optionally, the first determining module 804 is further configured to:

correspondingly, the first determination module 804 is further configured to:

extracting the characteristics of the user information by using a user characteristic extraction module to obtain user characteristics; extracting features of each candidate style in the first style dimension of the first candidate recommendation information by using a style feature extraction module to obtain style features of each candidate style; and according to the user characteristics and each style characteristic, calculating the conversion rate of each candidate style by using a calculation module.

Optionally, the first determining module 804 is further configured to:

according to the information content of the user information and at least one candidate recommendation information, carrying out validity screening on each candidate style in the first style dimension; and determining the target style in the first style dimension according to the screening result and the conversion rate.

Optionally, the second determination module 806 is further configured to:

based on target display templates and information contents of each candidate recommendation information, pre-trained recommendation information estimation models are utilized to estimate information conversion rates of the candidate recommendation information under each target display template, wherein the recommendation information estimation models are neural network models obtained through training according to the display templates and the information contents of the sample recommendation information and the historical conversion rates of the sample recommendation information; and ordering each candidate recommendation information according to the information conversion rate.

Optionally, the second determination module 806 is further configured to:

screening each candidate recommendation information according to a preset information online strategy to obtain screened candidate recommendation information; and sorting the screened candidate recommendation information according to the information conversion rate.

Optionally, the presentation module 808 is further configured to:

determining target recommendation information from at least one candidate recommendation information according to the sorting result; and displaying the target recommendation information.

Optionally, the acquisition module 802 is further configured to:

acquiring a plurality of initial recommendation information; and screening and obtaining at least one candidate recommendation information from the plurality of initial recommendation information according to a preset information screening strategy.

The above is a schematic scheme of a recommended information display device of this embodiment. It should be noted that, the technical solution of the recommended information display device and the technical solution of the recommended information display method belong to the same concept, and details of the technical solution of the recommended information display device, which are not described in detail, can be referred to the description of the technical solution of the recommended information display method.

FIG. 9 illustrates a block diagram of a computing device provided in accordance with an embodiment of the present application. The components of computing device 900 include, but are not limited to, memory 910 and processor 920. Processor 920 is coupled to memory 910 via bus 930 with database 950 configured to hold data.

Computing device 900 also includes an access device 940, access device 940 enabling computing device 900 to communicate via one or more networks 960. Examples of such networks include PSTN (Public Switched Telephone Network ), LAN (local area network), WAN (Wide Area Network, wide area network), PAN (Personal Area Network ), or a combination of communication networks such as the internet. The access device 940 may include one or more of any type of network interface, wired or wireless (e.g., NIC (Network Interface Controller, network interface card)), such as an IEEE802.12 WLAN (Wireless Local Area Networks, wireless local area network) wireless interface, wi-MAX (World Interoperability for Microwave Access, worldwide interoperability for microwave access) interface, ethernet interface, USB (Universal Serial Bus ) interface, cellular network interface, bluetooth interface, NFC (NearField Communication ) interface, and so forth.

In one embodiment of the present application, the above-described components of computing device 900 and other components not shown in FIG. 9 may also be connected to each other, for example, by a bus. It should be understood that the block diagram of the computing device illustrated in FIG. 9 is for exemplary purposes only and is not intended to limit the scope of the present application. Those skilled in the art may add or replace other components as desired.

Computing device 900 may be any type of stationary or mobile computing device including a mobile computer or mobile computing device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), mobile phone (e.g., smart phone), wearable computing device (e.g., smart watch, smart glasses, etc.), or other type of mobile device, or a stationary computing device such as a desktop computer or PC. Computing device 900 may also be a mobile or stationary server.

Wherein, the processor 920 implements the steps of the recommended information display method when executing the instruction.

The foregoing is a schematic illustration of a computing device of this embodiment. It should be noted that, the technical solution of the computing device and the technical solution of the recommended information display method belong to the same concept, and details of the technical solution of the computing device, which are not described in detail, can be referred to the description of the technical solution of the recommended information display method.

An embodiment of the present application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the recommended information presentation method as described above.

The above is an exemplary version of a computer-readable storage medium of the present embodiment. It should be noted that, the technical solution of the storage medium and the technical solution of the recommended information display method belong to the same concept, and details of the technical solution of the storage medium, which are not described in detail, can be referred to the description of the technical solution of the recommended information display method.

The foregoing describes specific embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The computer instructions include computer program code that may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth.

It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application. Further, those skilled in the art will also appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all necessary for the present application.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.

The above-disclosed preferred embodiments of the present application are provided only as an aid to the elucidation of the present application. Alternative embodiments are not intended to be exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the teaching of this application. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. This application is to be limited only by the claims and the full scope and equivalents thereof.

Claims

1. A recommended information display method, characterized by comprising:

estimating the conversion rate of each candidate style in a first style dimension of first candidate recommendation information, and determining a target style in the first style dimension according to the conversion rate, wherein the first candidate recommendation information is any one of the at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information;

generating a target display template of the first candidate recommendation information according to target styles of the first candidate recommendation information in each style dimension;

2. The method of claim 1, wherein estimating the conversion rate of each candidate style in the first style dimension of the first candidate recommendation information comprises:

3. The method of claim 2, wherein the conversion rate estimation model comprises a user feature extraction module, a style feature extraction module, and a calculation module;

the estimating, according to the user information, the conversion rate of each candidate pattern in the first pattern dimension of the first candidate recommendation information by using a pre-trained conversion rate estimation model includes:

extracting the characteristics of the user information by using the user characteristic extraction module to obtain user characteristics;

extracting features of each candidate style in the first style dimension of the first candidate recommendation information by using the style feature extraction module to obtain style features of each candidate style;

and according to the user characteristics and the style characteristics, calculating the conversion rate of each candidate style by using the calculation module.

4. A method according to any of claims 1-3, wherein said determining a target pattern in a first pattern dimension from said conversion rate comprises:

according to the user information and the information content of the at least one candidate recommendation information, carrying out validity screening on each candidate style in the first style dimension;

5. The method of claim 1, wherein the ranking the candidate recommendation information based on the target presentation template for the candidate recommendation information comprises:

based on target display templates and information contents of each candidate recommendation information, pre-trained recommendation information estimation models are utilized to estimate information conversion rates of the candidate recommendation information under each target display template, wherein the recommendation information estimation models are neural network models obtained through training according to the display templates and the information contents of sample recommendation information and the historical conversion rates of the sample recommendation information;

and sequencing each candidate recommendation information according to the information conversion rate.

6. The method of claim 5, wherein the ranking the candidate recommendation information according to the information conversion rate comprises:

and sorting the candidate recommendation information after screening according to the information conversion rate.

7. The method of claim 1, wherein the presenting of the recommended information according to the ranking result comprises:

Determining target recommendation information from the at least one candidate recommendation information according to the sorting result;

and displaying the target recommendation information.

8. The method of claim 1, wherein the obtaining at least one candidate recommendation information comprises:

acquiring a plurality of initial recommendation information;

9. A recommended information display device, characterized by comprising:

a first determining module configured to estimate a conversion rate of each candidate style in a first style dimension of first candidate recommendation information, and determine a target style in the first style dimension according to the conversion rate, wherein the first candidate recommendation information is any one of the at least one candidate recommendation information, and the first style dimension is any style dimension of the first candidate recommendation information;

10. A computing device comprising a memory, a processor, and computer instructions stored on the memory and executable on the processor, wherein the processor, when executing the instructions, implements the steps of the method of any of claims 1-8.

11. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 8.