CN113806622A

CN113806622A - Recommendation method, device and equipment

Info

Publication number: CN113806622A
Application number: CN202010535951.4A
Authority: CN
Inventors: 许家铭; 吴蒙蒙; 赵笑天; 郭城; 王治力; 刘慧红; 陈雯; 陈立; 龙志勇; 许海波; 安瑾; 潘明明
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2020-06-12
Filing date: 2020-06-12
Publication date: 2021-12-17

Abstract

The embodiment of the application provides a recommendation method, a recommendation device and recommendation equipment, wherein the method comprises the following steps: determining page information aiming at page display of an object to be recommended of a user, wherein the page information is used for indicating the style of a page for displaying the object to be recommended; and sending the page information to a terminal used by the user so that the terminal can display the page of the object to be recommended according to the page information. According to the method and the device, the diversity of the objects to be recommended displayed by the terminal is ensured based on the page information, and the attraction of the objects to be recommended to the user is favorably improved.

Description

Recommendation method, device and equipment

Technical Field

The present application relates to the field of internet technologies, and in particular, to a recommendation method, apparatus, and device.

Background

With the continuous development of internet technology, the amount of internet information grows exponentially, and in order to enable a user to obtain targeted information in time, personalized recommendation services for the user have been widely applied to various fields to automatically recommend related objects, such as commodities, audio, video and the like, to the user.

Generally, the server can determine an object to be recommended for the user, and send information of the object to be recommended to the terminal, and the terminal can display the information of the object to be recommended by adopting a fixed page template so as to realize page display for the object to be recommended, so that the user can know the information of the object to be recommended, and personalized recommendation for the user is realized.

Although the recommendation mode can achieve 'thousands of people and thousands of faces' of the recommended object for the user, the terminal adopts the fixed page template to display the information of the object, the page style used by the terminal to display the object to be recommended is single, and the attraction to the user is insufficient. Therefore, how to improve the attraction of page display to users becomes a problem to be solved urgently at present.

Disclosure of Invention

The embodiment of the application provides a recommendation method, a recommendation device and recommendation equipment, which are used for solving the problem that how to improve the attraction of page display to a user in the prior art is urgent to solve at present.

In a first aspect, an embodiment of the present application provides a recommendation method, where the method includes:

determining page information aiming at page display of an object to be recommended of a user, wherein the page information is used for indicating the style of a page for displaying the object to be recommended;

and sending the page information to a terminal used by the user so that the terminal can display the page of the object to be recommended according to the page information.

In a second aspect, an embodiment of the present application provides a recommendation method, where the method includes:

receiving page information sent by a server side, wherein the page information is determined by the server side aiming at the page display of an object to be recommended of a user, and the page information is used for indicating the style of a page for displaying the object to be recommended;

and performing page display on the object to be recommended according to the page information.

In a third aspect, an embodiment of the present application provides a recommendation device, including:

the system comprises a determining module, a judging module and a recommending module, wherein the determining module is used for determining page information aiming at page display of an object to be recommended of a user, and the page information is used for indicating a style of a page for displaying the object to be recommended;

and the sending module is used for sending the page information to a terminal used by the user so that the terminal can display the page of the object to be recommended according to the page information.

In a fourth aspect, an embodiment of the present application provides a recommendation device, including:

the system comprises a receiving module, a recommendation module and a recommendation module, wherein the receiving module is used for receiving page information sent by a server side, the page information is determined by the server side aiming at page display of an object to be recommended of a user, and the page information is used for indicating a style of a page for displaying the object to be recommended;

and the display module is used for displaying the page of the object to be recommended according to the page information.

In a fifth aspect, an embodiment of the present application provides a computer device, including: a memory, a processor; wherein the memory is configured to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of any of the first aspects.

In a sixth aspect, an embodiment of the present application provides a terminal, including: a memory, a processor; wherein the memory is configured to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of any of the second aspects.

Embodiments of the present application also provide a computer-readable storage medium storing a computer program, the computer program comprising at least one code, which is executable by a computer to control the computer to perform the method according to any one of the first aspect.

Embodiments of the present application also provide a computer-readable storage medium storing a computer program, the computer program comprising at least one code, which is executable by a computer to control the computer to perform the method according to any one of the second aspect.

Embodiments of the present application also provide a computer program, which is used to implement the method according to any one of the first aspect when the computer program is executed by a computer.

Embodiments of the present application also provide a computer program, which is used to implement the method according to any one of the second aspect when the computer program is executed by a computer.

According to the recommendation method, the device and the equipment provided by the embodiment of the application, page information is determined by page display of an object to be recommended of a user, the page information is used for indicating the style of a page displayed by the object to be recommended, the page information is sent to a terminal used by the user, so that the terminal can perform page display on the object to be recommended according to the page information, the page information displayed on the page of the object to be recommended can be determined in a targeted manner and sent to the terminal, the terminal can display the page of the object to be recommended based on the page information, the purpose of ensuring the diversity of the object to be recommended displayed by the terminal based on the page information is achieved, and compared with the case of performing page display on the object to be recommended by adopting a single page style, the attraction of the object to be recommended to the user is favorably improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a schematic view of an application scenario according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a recommendation method provided in an embodiment of the present application;

fig. 3 is a schematic flowchart of a recommendation method according to an embodiment of the present application;

FIG. 4 is a schematic diagram of page-level display optimization provided by an embodiment of the present application;

FIGS. 5A and 5B are schematic diagrams of element-level display optimization provided by an embodiment of the present application;

fig. 6 is a schematic diagram of obtaining materials based on a material service platform according to an embodiment of the present application;

fig. 7 is a schematic diagram of a network structure for implementing sequence-to-sequence conversion according to an embodiment of the present application;

FIG. 8 is a schematic structural diagram of a neural network according to an embodiment of the present application;

fig. 9 is a schematic diagram of pit bit region division according to an embodiment of the present application;

fig. 10 is a schematic diagram illustrating page display optimization based on user characteristics and an object to be recommended according to an embodiment of the present application;

11A-11D are schematic diagrams of various display manners corresponding to the same pair of recommended objects according to an embodiment of the present application;

fig. 12 is a schematic flowchart of a recommendation method according to another embodiment of the present application;

fig. 13 is a schematic structural diagram of a recommendation device according to an embodiment of the present application;

FIG. 14 is a schematic structural diagram of a computer device according to an embodiment of the present application;

fig. 15 is a schematic structural diagram of a recommendation device according to another embodiment of the present application;

fig. 16 is a schematic structural diagram of a terminal according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The terminology used in the embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the examples of this application 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, and "a" and "an" typically include at least two, but do not exclude the presence of at least one.

It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.

The words "if", as used herein, may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to a detection", depending on the context. Similarly, the phrases "if determined" or "if detected (a stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when detected (a stated condition or event)" or "in response to a detection (a stated condition or event)", depending on the context.

It is also noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a good or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such good or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a commodity or system that includes the element.

In addition, the sequence of steps in each method embodiment described below is only an example and is not strictly limited.

For the convenience of those skilled in the art to understand the technical solutions provided in the embodiments of the present application, a technical environment for implementing the technical solutions is described below.

The recommendation method commonly used in the related art mainly comprises the steps that a server determines an object to be recommended for a user, information of the object to be recommended is sent to a terminal, the terminal displays the object to be recommended by adopting a fixed page template, the page style used by the terminal for displaying the object to be recommended is single, and the attraction to the user is insufficient. Therefore, there is a need in the related art for a recommendation method that can improve the attraction of the object to be recommended to the user.

Based on the actual technical requirements similar to those described above, the recommendation method provided by the application can improve the attraction of the object to be recommended to the user by using a technical means.

The following describes a recommendation method provided in various embodiments of the present application in detail through an exemplary application scenario.

As shown in fig. 1, the terminal 11 may send a request message to the server 12 according to an input of the user to request the server 12 to return an object (hereinafter, referred to as an object to be recommended) to be recommended to the user to the terminal 11, where the request message may be used to indicate an identifier of the user to implement personalized recommendation for the user, and of course, in other embodiments, the request message may also be used to indicate other information, for example, a current network type of the terminal 11, an operating system type of the terminal 11, and the like, which is not limited in this application. The terminal 11 may be, for example, a mobile phone, a tablet Computer, a Personal Computer (PC), etc., and of course, in other embodiments, the terminal 11 may be in other forms, which is not limited in this application.

After the server 12 receives the request message, on the one hand, the server 12 may determine the object to be recommended for the user according to the request message. The number of the objects to be recommended may be one or more, and the objects to be recommended may specifically be a list of the objects to be recommended in a plurality of cases. On the other hand, the server 12 may determine page information for page display of the object to be recommended of the user, and send the page information to the terminal 11, so that the terminal 11 may perform page display on the object to be recommended according to the page information.

It should be noted that, for different application scenarios, the types of the objects to be recommended may be different, for example, the objects to be recommended may specifically be the commodities to be recommended in an e-commerce scenario, the objects to be recommended may specifically be the advertisements to be recommended in an advertisement promotion scenario, and the objects to be recommended may specifically be the videos to be recommended in a video playing scenario.

After determining the page information, the server 12 may send the page information to the terminal 11, and the terminal 11 may perform page display on the object to be recommended according to the page information.

It should be noted that, in fig. 1, the server 12 is taken as an example to determine the object to be recommended and the page information, it may be understood that, in other scenarios, the object to be recommended and/or the page information may be determined by other devices besides the server 12, and the devices that determine the page information of the object to be recommended and the page information may be the same device or different devices.

For convenience of description, the following description mainly takes object recommendation in an application as an example, where a front end of the application is deployed in the terminal 11, and a server of the application is deployed in the server 12. As shown in fig. 2, the server may invoke the recommendation module based on a request of the front end to determine an object to be recommended, and the server may invoke an intelligent User Interface (UI) module to determine page information. The recommendation module is a functional module for providing and determining an object to be recommended, and the intelligent UI module is a functional module for providing and determining page information. The recommendation module and the intelligent UI module may be implemented by the server 12 or by other devices than the server 12.

Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. The embodiments described below and the features of the embodiments can be combined with each other without conflict.

Fig. 3 is a flowchart illustrating a recommendation method according to an embodiment of the present application, where an execution subject of the embodiment may be a device on a server side, for example, the server 12 in fig. 1 or another device other than the server 12. As shown in fig. 3, the method of this embodiment may include:

step 301, determining page information for page display of an object to be recommended of a user, wherein the page information is used for indicating a style of a page for displaying the object to be recommended;

step 302, sending the page information to a terminal used by the user, so that the terminal can display the page of the object to be recommended according to the page information.

In this embodiment of the application, optionally, the presentation of the page may be optimized based on the template granularity, that is, page-level presentation optimization is implemented, and based on this, step 301 may specifically include: according to the user characteristics of the user, selecting one page template from a plurality of preset page templates as a target page template to obtain the page information, wherein the target page template corresponds to a preset page style, and the page information is specifically used for indicating the target page template. The page information may include, for example, an Identifier (ID) of the target page template, and certainly, in other embodiments, the target page template may be indicated to the terminal in other manners, which is not limited in this application.

In the page-level display optimization, the intelligent UI module may, for example, use the click rate of the page overall template as an optimization target, and perform algorithm learning and optimization by using the user characteristics and the template characteristics, and certainly, in other embodiments, the intelligent UI module may also use other indexes of the page overall template as an optimization target, which is not limited in this application. The intelligent UI module and the recommendation module are relatively independent, and the implementation mode can be as shown in FIG. 4. Referring to fig. 4, the server may obtain the object to be recommended by calling the recommendation module and the template ID (i.e., UI _ ID) of the target page template by calling the intelligent UI module under the trigger of receiving the request message of the front end, and return the object to be recommended and the UI _ ID to the front end, so that the terminal 11 performs page rendering using the UI _ ID and performs page content filling using the object to be recommended.

It should be noted that, in fig. 4, the server calls the recommendation module through the request message as an example, and of course, in other embodiments, the server may also call the recommendation module in other manners, which is not limited in this application. In fig. 4, the server invokes the intelligent UI module through the user static characteristics (for example, the ID of the user, the current network type of the terminal 11, the type of the operating system of the terminal 11, and the like) determined based on the request message, and in other embodiments, the server may invoke the intelligent UI module in other manners.

In an embodiment, the intelligent UI may obtain, based on the obtained static features of the user, dynamic features of the user, such as age, crowd, preference, and the like, based on the ID of the user, and further, the intelligent UI may predict, according to the static features of the user and the dynamic features of the user, a preset index, such as a click rate, of the user for each page template, and may use a page template with a highest click rate as a target page template.

For example, the plurality of page templates may include a single-column template, a double-column template and a three-column template according to the number of columns of objects in one page, where a single-column template may display one column of objects in the same page, a double-column template may display two columns of objects in the same page, and a three-column template may display three columns of objects in the same page.

In the embodiment of the application, the following four problems of page-level display optimization are considered, and an element-level display optimization mode is further provided on the basis of the page-level display optimization. First, since the user's feedback on an object is closely related to the object (e.g., a commodity), page-level display optimizes this way of optimizing the stripped object, and is not targeted to the object. Second, the presentation varies widely from one template to another, resulting in a large variation in the final feedback situation. An obvious example is that a three-column template has natural advantages in independent visitor Click-through Rate (UVCTR) due to the larger number of objects presented per screen, while a single-column template has obvious advantages in PVCTR due to the smaller number of objects presented per screen. This difference, which has an impact on the indicator, severely interferes with the algorithm's learning for the visual presentation. Third, for the same user, page-level display optimization often returns a single page style, thereby reducing the browsing experience of the user. And fourthly, a plurality of page templates for page level display optimization depend on preset, the number is limited, and from the design of one page template to joint debugging and online, the flow is complex and the period is long, so that the mode of the page template has large workload and is not beneficial to the iteration of the algorithm.

Based on the above problems, the embodiment of the application further breaks through the template granularity, allows the combination of the page elements, and makes a display decision based on a plurality of candidate materials of the object to be recommended to determine the materials (marked as the page materials) that can be displayed by the object to be recommended. Based on this, step 301 may specifically include: selecting partial candidate materials meeting certain conditions from the multiple candidate materials of the object to be recommended as page materials to be displayed, wherein the page materials correspond to page elements, and the page elements correspond to preset element styles; and generating the page information based on the page material, wherein the page information is specifically used for indicating the page material and the corresponding page elements thereof.

The page information can be combined with a dynamically compatible basic page template, and the basic page template can support free combination of page elements, so that a decision result of making a display decision based on a plurality of candidate materials of an object to be recommended can be presented on the dynamically compatible basic template. Therefore, hundreds of display styles can be obtained, so that a user can browse various display styles aiming at different objects to be recommended in the same page, and the browsing experience of the user can be ensured as much as possible.

In element-level presentation optimization, one implementation may be as shown in FIG. 5A. Referring to fig. 5A, the server may obtain the object to be recommended by calling the recommendation module under the trigger of receiving the front-end request message, call the intelligent UI module based on the object to be recommended to determine the page material to be displayed and the page element corresponding to the page material, and indicate the page element to be displayed and the page material corresponding to the page element to be displayed to the front end, so that the front end dynamically renders based on the page material to be displayed and the page element corresponding to the page material on the basis of the basic page template.

Alternatively, to accommodate different recommendation scenarios, another implementation may be as shown in fig. 5B. Referring to fig. 5B, the server may call the compatible module under the trigger of receiving the request message from the front end, the compatible module obtains an object to be recommended by calling the recommendation module a and/or the recommendation module B, and calls the intelligent UI module based on the object to be recommended to determine a page material to be displayed and a page element corresponding to the page material to be displayed and return the page material to the server, and the server finally indicates the page element to be displayed and the page material corresponding to the page material to be displayed to the front end, so that the front end dynamically renders based on the page material to be displayed and the page element corresponding to the page material on the basis of the basic page template. The compatible module is configured to provide a function of being compatible with multiple recommendation modules, and it should be noted that, taking the number of recommendation modules as 2 in fig. 5B as an example, different recommendation modules may correspond to different recommendation scenarios.

In the embodiment of the application, under the condition that a plurality of recommendation scenes need to be compatible, a plurality of candidate materials of the object to be recommended can have a corresponding relation with the recommendation scenes, namely, display decisions can be made based on different candidate materials under different recommendation scenes. Based on this, before selecting partial candidate materials from the multiple candidate materials of the object to be recommended as page materials to be displayed, the method may further include: and acquiring the candidate materials based on the recommendation scene corresponding to the object to be recommended.

Optionally, the candidate materials may be obtained by calling an obtaining interface provided by the material service platform. The material service platform can provide a query function of candidate materials. For example, the material service platform may be called by using the identifier of the object to be recommended as an entry parameter, so that the material service platform may acquire and return a plurality of candidate materials of the object to be recommended. For another example, the material service platform may be called by using the identifier of the object to be recommended and the identifier of the recommendation scene as entry parameters, so that the material service platform may acquire and return a plurality of candidate materials of the object to be recommended in a specific recommendation scene. Of course, in other embodiments, the material service platform may also be called in other manners, which is not limited in this application.

As shown in fig. 6, based on different material obtaining manners, the material service platform may, for example, divide the material into at least two types of materials, such as a first type of material and a second type of material, where the first type of material may be obtained through the service interface 1, and the second type of material may be obtained through the service interface 2. After the material service platform obtains the candidate materials corresponding to the object to be recommended, the further algorithm module may select a part of the candidate materials from the plurality of candidate materials of the object to be recommended as the page materials to be displayed, and generate the page information based on the page materials. As can be seen from fig. 6, the algorithm module does not sense different acquisition modes of the material any more through the material service platform, which is beneficial to reducing the implementation coupling.

In the embodiment of the present application, the candidate material may include a picture material. Based on this, the candidate material may specifically include original picture material and generated picture material derived based on the original picture material to provide rich picture candidates.

In order to improve the convenience of setting the original picture material by the provider of the object to be recommended, the original picture material may be uploaded by the provider of the object to be recommended through an object provider platform. Wherein the object provider platform can provide user interaction functionality. It should be noted that, for different application scenarios, the providers of the objects to be recommended may be different, for example, the provider of the object to be recommended for an e-commerce scenario may be a merchant, and the provider of the object to be recommended for a video playing scenario may be a video publisher.

In an embodiment, the method provided in the embodiment of the present application may further include: and presetting the original picture material, and taking the picture obtained after the presetting as a generated picture material. For example, the pre-setting process may include one or more of a picture size adaptive cropping process for a single raw picture material, a creative picture generation process for a single raw picture material, or a stitching process for multiple raw picture materials.

In the picture size adaptive cropping process, the object body may be recognized first, then a portion including the object body is cropped, and finally the portion is scaled or expanded to a suitable size. Therefore, the object body can be highlighted, and an optimized space can be provided for the space utilization rate of the page. In the creative picture generation process, image processing technologies such as main body recognition, edge detection, style migration and the like can be utilized to enrich picture materials.

In the embodiment of the application, in order to improve the participation degree of the provider in the generation of the picture material of the object to be recommended, optionally, the provider may participate in the generation process of the picture material. Based on this, the embodiment of the present application may include: and presetting the original picture material, and displaying the picture obtained after the presetting to the provider through the object provider platform so that the provider modifies or confirms the picture through the object provider platform to obtain the generated picture material. In the object provider platform, a user can acquire a picture obtained after the server side performs preset processing on the original picture material, and the picture is modified or confirmed through the object provider platform to obtain a final generated picture material.

Optionally, in order to enable the provider to know the feedback conditions of different picture materials, the embodiment of the present application may further include: and acquiring user feedback statistics aiming at the original picture material and the generated picture material respectively, and prompting the user feedback statistics to the provider through the object provider platform. The feedback of the user to the material may be, for example, clicking, converting, and the like, and the user feedback statistics of the user to the original picture material and the generated picture material may be obtained based on the log of the user.

In the embodiment of the present application, the candidate material may include a text material. Based on this, the method provided in the embodiment of the present application may further include: and obtaining the text material based on the object text of the object to be recommended so as to improve the intelligence of obtaining the text material. The object text of the object to be recommended may be provided by a provider of the object to be recommended, for example, and of course, the object text may be obtained in other ways in other embodiments, which is not limited in this application. The object text may include, for example, marketing benefits, etc., and in other embodiments, the object text may include other contents, which is not limited in this application.

In one embodiment, the text material may be obtained based on Named Entity tags obtained based on Named Entity Recognition (NER) technology. Taking an e-commerce scenario as an example, the named entity tag may include, for example, a brand, a style, a material, a category, and the like, and certainly, the named entity tag may also be in other forms in other embodiments, which is not limited in this application. Considering that the part-of-speech tags are a language-level classification system, the method cannot be well adapted to the actual recommendation scene, for example, although some meaningless names are not suitable for presentation, the part-of-speech tags "nouns" cannot be directly shielded, and the meaningless words cannot be effectively identified from the "nouns". However, the named entity tag obtained based on the NER technology can be subjected to text labeling aiming at the actual recommendation scene, so that the named entity tag can be matched with the actual recommendation scene, and the problem that the part-of-speech tag is not matched with the actual recommendation scene is avoided.

Based on this, the obtaining of the text material based on the object text of the object to be recommended may specifically include: determining a plurality of front words of which the named entity tags belong to the named entity tag set and the occurrence rate is high from the object text of the object to be recommended based on a preset named entity tag set; and extracting short texts containing the words from the long texts of the objects to be recommended. In the text 1, the text content of the text 2 is extracted, so that the length of the text 2 is shorter than that of the text 1, the text 1 is a long text relative to the text 2, and the text 2 is a short text relative to the text 1.

The long text may be a long title, the corresponding short text may be a short title, the long text may be a long reason for recommendation, and the corresponding short text may be a short reason for recommendation. Of course, in other embodiments, the long text may also be other types of information, which is not limited in this application.

It is contemplated that the named entity labels that are emphasized may be different in different scenarios, such as where the brand is a key named entity label in a top page recommendation scenario, and where the brand is a general knowledge, and where style, texture, etc. are key named entity labels in an in-store recommendation scenario. Thus, the set of named entity tags can correspond to a recommendation scenario. Taking a title scene as an example, the most basic requirement is to indicate what an object is, so a named entity tag set corresponding to a long title may include "category", and other named entity tags included in the named entity tag set may be implemented according to requirements.

Optionally, the named entity tag may belong to a plurality of previous words with high occurrence rate in the named entity tag set based on a Term Frequency-Inverse Document Frequency (TF-IDF) algorithm. Specifically, the object text of the object to be recommended may be used as a file, a TF-IDF weight value of each word is calculated for the file by using a TF-IDF algorithm, and a TF-IDF weight value TOPN of a named entity tag belonging to a preset named entity tag set is determined based on the TF-IDF weight value of each word, so as to obtain a plurality of words.

Or, optionally, determining that the named entity tag belongs to the named entity tag set and is a plurality of words with high occurrence rate based on the TextRank algorithm. Specifically, for an object text of an object to be recommended, operations such as word segmentation and word stop can be performed first, then the words are used as nodes, a co-occurrence relationship between the words is an edge composition, interface weight is initialized, iterative solution is performed, and finally the TextRank weight value of the words is obtained. Similar to TF-IDF, the TF-IDF weight value TOPN of a named entity tag belonging to a preset named entity tag set can be determined based on the weight value of each word TextRank.

Considering that the object text is usually short, the frequency TF of the words in the TF-IDF algorithm is small, the calculation of the values of the words TF-IDF mainly depends on the frequency IDF of the inverse documents and tends to global information, and the modeling of the TextRank algorithm on the object text is focused on the local information of the object text.

In order to avoid that the generated material text includes an undesired word, the extracting a short text including the plurality of named entity tags from the long text of the object to be recommended may specifically include: and extracting short texts which contain the plurality of named entity labels and do not contain words in a preset negative word list from the long texts of the objects to be recommended.

Illustratively, the extraction of short text from long text may be implemented based on a sequence-to-sequence conversion algorithm. Taking an example of a sequence-to-sequence conversion algorithm realized based on a Pointer Network (Pointer Network), the embodiment of the present application provides a text generation method based on multivariate Positive and Negative word list control, namely MPNV-Pointer Network (Multi Positive Negative voice Pointer Generator Network), and the text generation method can provide the following three-aspect capabilities:

(1) a forward vocabulary, which may be based on a Multi-Source Pointer Network (Multi-Source Point Network) extension, to add the aforementioned determined plurality of words to a Multi-Source information Encoder (Encode) for algorithmic production. The formula expression of the model algorithm of the encoder may be as shown in formula (1).

P(y_t|x,y<t)＝λP_src+(1-λ)P_msFormula (1)

Wherein x represents encoder input text; y denotes a decoder (decoder) output text; t represents the tth step of the decoder output; y is_tRepresenting the output of the decoder at step t, P (y)_t|x，y<t) denotes the t-th output y of the decoder at a given input x and the first t-1 output in the t-th step₁,..,y_t-1]Prediction of y_tProbability distribution over the entire vocabulary of model inputs; p_srcProbability distribution, P, represented on source text vocabulary_msRepresenting a probability distribution over the forward vocabulary; lambda represents the fusion weight and has a value range of [0,1 ]]；P_ms、Context_msAnd λ can be obtained using the following formula (2) and formula (3), respectively.

P_ms,Context_ms＝Attention(d_t,[EMB₁,EMB₂,……，EMB_n]) Formula (2)

Wherein, Context_msA vector representation representing a forward vocabulary; p_msRepresenting a probability distribution over the forward vocabulary; d_tRepresenting the vector expression of the t step of the decoder; EMB₁To EMB_nEmbedding representing each word in the forward vocabulary; attention means that Attention (Attention) mechanism calculation is performed to obtain probability distribution and vector expression of fusion context.

λ＝Gate(Context_src,Context_ms,d_t) Formula (3)

Wherein the content of the first and second substances,

representing source textA vector expression of (a); context_msA vector representation representing a forward vocabulary; d_tRepresenting the vector expression of the t step of the decoder; and the Gate represents a Gate mechanism, the probability distribution of the source text and the forward word list is fused, the vector expression of the source text, the vector expression of the forward word list and the vector expression of the t step of the decoder are input, and the fusion probability is output.

(2) The negative word list can be used in a model Inference (reference) stage, and the distribution of the word list is controlled by a mask or the like so as to achieve the purpose of shielding the specified words, and the formula expression of the negative word list can be shown as the following formula (4).

Wherein the content of the first and second substances,

the output of the decoder after the t step is subjected to negative control is shown;

and (4) representing the probability distribution of the decoder on the whole vocabulary input by the model after the t step is subjected to negative control. P (y)_t|x，y<t) denotes the t-th output y of the decoder at a given input x and the first t-1 output in the t-th step₁,..,y_t-1]Prediction of y_tProbability distribution over the entire vocabulary of model inputs;

whether the ith word on the word list is a negative word or not is shown, 0 is a negative word, and 1 is a non-negative word; v represents the entire vocabulary of model inputs.

(3) Variable length control, which can be at the model bundle/column Search (Beam Search) stage to retain all results that satisfy the length interval. Specifically, a plurality of short texts can be determined when determining the word in each step, so that a plurality of combinations can be obtained after combining with the previous step, and a plurality of short texts can be finally obtained, so that the short texts meeting the length interval can be determined from the plurality of short texts finally obtained.

The structure of the MPNV-Pointer Network may be as shown in fig. 7, where the source text src in the vector form in reference to fig. 7 may be input as one path of the encoder, the forward word list ms in the vector form may be input as another path of the encoder, and both the source text in the vector form and the forward word list of the vector elements may include a plurality of vector elements, and one vector element may correspond to one word. H1 to hn in the encoder of fig. 7 can be implemented as a Long Short-Term Memory network (LSTM), ht (t takes 0,1 … … n) can represent the hidden layer corresponding to the t step of the source text in the encoder; z1 to zn can be implemented as another bidirectional LSTM, and zt (t takes 0,1 … … n) can represent the hidden layer corresponding to each t step of the forward vocabulary in the encoder; h1 and h2 in the decoder can be implemented as one-way LSTM, ht (t takes 0 and 1 … … n) represents the hidden layer corresponding to the output t step in the decoder. Specifically, the source text x in the encoder goes through a bi-directional LSTM and then sums d_tComputing the Attention mechanism to obtain the Context_srcAnd P_src(ii) a The forward vocabulary in the encoder goes through the bi-directional LSTM and then d_tComputing the Attention mechanism to obtain the Context_msAnd P_ms(ii) a The t step of the decoder is processed by LSTM to obtain d_tIt and Context_src、Context_msA join-gate mechanism, fuse P_msAnd P_srcTo obtain a word distribution P (y)_t|x,y<t). If there is a mask, then multiply the mask vector to get the word distribution

In this embodiment of the application, optionally, in order to improve flexibility of candidate elements of an object to be recommended, at least some candidate elements of the object to be recommended may be dynamically specified. Based on this, the selecting partial candidate materials from the multiple candidate materials of the object to be recommended as the page materials to be displayed includes: receiving a trigger message, wherein the trigger message is used for indicating candidate materials corresponding to at least one page element respectively; the candidate elements are at least partial candidate elements of the object to be recommended; and under the trigger of the trigger message, selecting partial candidate materials from the candidate materials of the object to be recommended as page materials to be displayed. The trigger message may be, for example, a module-to-smart UI module compatible message in fig. 5B.

After determining a plurality of candidate materials of the object to be recommended, any one of the following first to third ways may be adopted to select a part of candidate materials from the plurality of candidate materials as page materials to be displayed.

In the first mode, the display decision of the candidate material without distinguishing the user or the crowd may specifically include:

determining a material score of each candidate material based on feedback conditions of different users for each candidate material of the object to be recommended; and according to the material scores of the candidate materials, determining the candidate materials with the scores meeting certain conditions from the candidate materials as the page materials to be displayed.

In one approach, the score of the candidate material may be determined, for example, based on a bayesian (Bandit) algorithm of Thompson sampling. The Bandit algorithm can be used for a very straightforward idea to solve the "exploration and utilization" (explicit-explicit) problem: how to obtain the maximum cumulative revenue (reward). It is assumed that the rewarded (in some scenarios, the benefit may be, for example, a click) generated by each material follows a probability distribution, which may be specifically Beta-Bernoulli Bandit. Let θ represent the mean value of rewarded corresponding to each material, and the rewarded corresponding to each selection of the material is Bernoulli distribution with θ as a parameter. Assuming that the prior distribution of θ is a Beta distribution, then its posterior distribution is also a Beta distribution. Based on the above, Beta sampling under respective parameters can be performed on the candidate materials of the object to be recommended, so as to obtain the material scores of the candidate materials.

Taking the example that the score of the candidate material is determined based on the Bandit algorithm, and the candidate material meeting a certain condition is the material with the largest score. The formula expression of the mode one algorithm can be expressed by, for example, the following formula (5).

Wherein k represents the kth material, ipv_kIndicating the number of clicks, pv, of the kth material_kAnd representing the exposure times of the kth material, Beta representing a Beta function, and choice representing the material with the maximum estimated click probability.

The first way only takes the feedback statistical information into consideration when determining the material score, and does not take into account other useful characteristics and information available in the recommendation scene, thereby further providing the following second way.

In a second mode, distinguishing the crowd for making the display decision of the candidate material specifically may include:

determining a material score of each candidate material based on a feedback condition of a target population for each candidate material of the object to be recommended, wherein the target population is a population to which the user belongs; and according to the material scores of the candidate materials, determining the candidate materials with the scores meeting certain conditions from the candidate materials as the page materials to be displayed.

For example, the material score of each candidate material may be determined based on a linear model trained in advance, wherein the linear model may be obtained based on historical feedback conditions for the target material in different people. Based on this, in an embodiment, the determining the material score of each candidate material based on the feedback condition of the target group for each candidate material of the object to be recommended may specifically include: and respectively inputting the crowd characteristics of the target crowd and the material characteristics of the candidate materials of the object to be recommended into a pre-trained linear model to obtain the material score of each candidate material.

In the second mode, for example, the score of the candidate material may be determined based on a Crowd (Crowd) Contextual Bayesian (CB) algorithm of LinUCB, so as to predict the click probability distribution of the Crowd on the material based on the Crowd characteristics, and explore and deliver different materials of the same object by using an optimistic estimation of an upper bound (upper bound).

Illustratively, an online log may be collected and recorded as (f)_i，u_j，r_ij) Wherein f is_iRepresenting the demographic characteristics of the population i, u_jRepresenting material j, r_ijRepresenting the feedback situation of the users in the crowd i to the material j (1 may for example represent a click, and 0 may for example represent an uncheck). According to the assumption of LinUCB, reward is modeled using a linear model,

wherein

Presentation material u_jThe parameter (c) of (c). After collecting the online logs, the parameters of the material may be updated using the following equations (6) and (7):

wherein the content of the first and second substances,

and

are all model parameters, wherein

Represents u_jThe covariance matrix of (a) is determined,

a reward vector representing material j.

In the on-line decision making, the rewarded prediction can be performed according to the estimated parameters, and in one embodiment, the rewarded prediction value and the uncertainty can be considered at the same time, which can be specifically shown in the following formula (8).

Wherein λ and α represent regularization parameters of the linear model and weighting parameters explored and utilized by the balance model, respectively, and the two parameters can be selected based on the results of the on-line AB test; i represents an identity matrix;

representing a reward mean estimate;

representing a reward confidence interval estimate.

The method of the mode of the two-linear model has the advantages of small model consumption, easy deployment, fast feature updating iteration and fast training speed, but relates to a large amount of matrix operations, and has lower supportable feature dimension and limited fine-grained depicting capability of users and materials. In addition, although both the Bandit algorithm and the contextal Bandit algorithm can achieve significant effects on intelligent UI material projection, it is difficult to model a user more effectively using only a linear model because the decision process of the user is often a complex and highly nonlinear process. Based on the above, the following third mode is further provided to more effectively utilize the features in the recommended scene and model the nonlinear rewarded, so as to obtain better generalization capability.

In a third mode, distinguishing a user to make a display decision of the candidate material specifically may include: determining a material score of each candidate material based on the feedback condition of the user for each candidate material of the object to be recommended; and according to the material scores of the candidate materials, determining the candidate materials with the scores meeting certain conditions from the candidate materials as the page materials to be displayed.

In one embodiment, a model can be used to model the multivariate polymalty interrelationships between user-object-materials, characterizing the fine-grained interest of the user. For example, the material placement problem in the smart UI can be modeled as a Rank problem, i.e., ordering the selectable elements for maximizing PVCTR or conversion Rate (CVR) and the like, given the user and the object.

For example, the material scores of the candidate materials may be determined based on a pre-trained neural network model, wherein the neural network model may be obtained based on historical feedback of different users for the target materials. Based on this, in an embodiment, the determining the material score of each candidate material based on the feedback condition of the user for each candidate material of the object to be recommended may specifically include: and respectively inputting the user characteristics of the user, the object characteristics of the object to be recommended and the material characteristics of each candidate material of the object to be recommended into a pre-trained neural network model to obtain the material score of each candidate material.

The user characteristics may include one or more of a user id, an age, a gender, and the like, and of course, the user characteristics may be other types of characteristics in other embodiments, which is not limited in this application. The object feature may include one or more of an object id, a category id, a pit id, a provider type (for an e-commerce scenario, the provider type may specifically be a merchant type, such as a platform 1 merchant or a platform 2 merchant), and the like, and of course, in other embodiments, the object feature may also be another type of feature, which is not limited in this application. The material characteristics may include, for example, one or more of a material id, a visually displayed characteristic, a visually hidden characteristic, and the like. Taking a picture as an example, the display characteristics may include one or more of color, brightness, contrast, whether a human face is included, layout, and the like, for example. Implicit features may include, for example, one or more of a sentence vector (content embedding) of the material, a sub-vector (word embedding) of the material, a cnn vector (cnn embedding) of the material, and the like.

Optionally, the neural network model comprises a first sub-network and a second sub-network; the first sub-network is used for determining pairwise association relations among the user characteristics, the object characteristics and the material characteristics; and the second sub-network is used for fusing the pairwise associations to obtain a final association relation, and the final association relation is used for representing the material score. Therefore, the material scores determined by the neural network model can fully take the interrelationship between the three types of features into consideration.

The structure of the neural network may be as shown in fig. 8, for example. Referring to fig. 8, in this model, the input may be divided into three feature groups, a user feature group, an object feature group, and a material feature group. First, the Embedding transformation can be performed on the features in each group, wherein id class features can be directly subjected to the Embedding operation, and numerical class features (such as the image CNN Embedding) can be directly converted into vectors. After obtaining the Embedding of each group feature, carrying out Embedding splicing on the features of each group to form 3 spliced vector representations. Then, pairwise cross prediction can be performed on the 3 spliced vectors, and modeling is performed on the user-object, the user-material and the object-material respectively. Then, the model can be trained to learn the association relationship among the three groups of features, so as to learn the pairwise association relationship. Finally, the three sets of vector predictors can be weighted using attention-based weighting to finally output a final association.

It should be noted that, in the specific description of the first to third modes, how to determine the material scoring for the material of the object to be recommended is mainly described, and after determining the material scoring for each of a plurality of candidate materials of the object to be recommended, the specific condition for selecting the page material from the candidate materials can be flexibly implemented according to the requirement. For example, the material with the material score greater than or equal to the score threshold may be used as the page material to be displayed, and of course, in other embodiments, the condition that the page material needs to meet may also be in other forms, which is not limited in this application.

In order to improve the flexibility of material selection control, the aforementioned trigger message may also be used to indicate a selection rule of page elements in a page, and the selection rule may specify, for example, which page elements are mandatory page elements, which page elements are optional page elements, and the like. Because the page elements correspond to the page materials, the material display decision can be carried out based on the selection rule of the page elements so as to obtain the page materials to be displayed. Based on the above, the candidate materials meeting a certain condition include materials of which the corresponding page elements meet the requirement of the selection rule.

Considering new materials which are online, the feedback information of the user to the new materials is limited, so that the material score of the new materials is low, and in order to avoid the problem that the new materials cannot be selected as the materials to be displayed and cannot take effect due to the low material score of the new materials, under the condition that the plurality of candidate materials include the new materials which are online within the latest preset time, part of the candidate materials are selected from the plurality of candidate materials of the object to be recommended as the page materials to be displayed, the method specifically includes: and selecting the new material from the candidate materials of the object to be recommended as a page material to be displayed according to a certain probability. The latest preset time period may be, for example, the latest 5 hours, and certainly, the latest preset time period may also be other values in other embodiments, which is not limited in this application. Taking the example of selecting a new material as a page material from a plurality of candidate materials of an object to be recommended according to a probability of 10%, the new material can be selected as the page material 10 times among the material selections from the plurality of candidate materials of the object to be recommended 100 times.

In the embodiment of the application, after the page materials to be displayed are determined, the page information can be generated based on the page materials. Under the condition that the material combination mode obtained based on the page material to be displayed is single, the page information can be directly generated by the material combination mode.

Or, optionally, in order to obtain better page information, a page material capable of obtaining multiple material combination modes may be determined first, then multiple material combination modes of the page material are determined according to the page material, finally, one material combination mode is selected from the multiple material combination modes, and page information is generated according to the selected material combination mode. For example, the page information may be generated by selecting a material combination mode with the largest sum of the material scores in the multiple material combination modes, but in other embodiments, the material combination mode may be selected from the multiple material combination modes in other modes, which is not limited in the present application.

As shown in fig. 9, the pit locations for presenting the object in the page may include, for example, an image area, a title area, a subtitle area, and a price area. Different areas in the pit positions correspond to different page element image areas and can be used for displaying images or video materials of the objects, the title area can be used for displaying titles of the objects, and the subheading area can be used for displaying text materials of the objects aiming at logistics services, text materials aiming at functional characteristics or text materials aiming at marketing rights and the like.

Taking the pit position of the object in the page as shown in fig. 9, and determining the page material of the object to be presented in the second or third manner as an example, as shown in fig. 10, picture selection may be performed based on the user characteristic and the object to be recommended, so as to select a picture or a video material to be presented from candidate materials of the object to be recommended, and recommendation reason selection may be performed based on the user characteristic and the object to be recommended, so as to select a text material to be presented from candidate materials of the object to be recommended for presentation in the subtitle region. In fig. 10, the page materials corresponding to the title area and the price area are fixed as an example.

After the materials to be displayed are obtained, material combination may be performed based on the materials to be displayed, so as to obtain a plurality of material combination modes, and taking the number of the material combination modes as 4 as an example, the display modes of the pit positions corresponding to the object to be recommended in the obtained plurality of material combination modes may be, for example, as shown in fig. 11A to 11D. The 4 display modes of fig. 11A to 11D may correspond to the 4 material combination modes one to one, and based on the 4 display modes of fig. 11A to 11D, it can be seen that the user can see different display modes of the same object to be recommended in different material combination modes. Finally, one material combination mode can be selected from the multiple material combination modes to generate page information, and as shown in fig. 10, assuming that the material combination mode corresponding to fig. 11A is selected from the multiple material combination modes, the page information can be generated based on the material combination mode corresponding to fig. 11A and sent to the terminal, and the terminal can present the object to be recommended to the user based on the page information in the mode of fig. 11A.

It should be noted that fig. 11A-11D take the object to be recommended as a washing machine as an example, where "high efficiency energy saving" is a text material for the functional features, and ". dot.automatic washing machine" is a title,

is the price of the item to be purchased,

is a text material for marketing rights and "next day is a text material for logistics services.

According to the recommendation method provided by the embodiment of the application, the page information is determined by the page display of the object to be recommended of the user, the page information is used for indicating the style of the page displayed by the object to be recommended, and is sent to the terminal used by the user, so that the terminal can display the page of the object to be recommended according to the page information, the page information displayed by the page of the object to be recommended can be determined in a targeted manner and is sent to the terminal, the terminal can display the page of the object to be recommended based on the page information, the diversity of the object to be recommended displayed by the terminal is ensured based on the page information, and compared with the page display of the object to be recommended by adopting a single page style, the attraction of the object to be recommended to the user is favorably improved.

Fig. 12 is a flowchart illustrating a recommendation method according to another embodiment of the present application, where an execution subject of the embodiment may be the terminal 11 in fig. 1. As shown in fig. 12, the method of this embodiment may include:

step 121, receiving page information sent by a server side, wherein the page information is determined by the server side for page display of an object to be recommended of a user, and the page information is used for indicating a style of a page for displaying the object to be recommended;

and step 122, performing page display on the object to be recommended according to the page information.

In the embodiment of the application, the terminal can support the capability of performing page display on the object to be recommended according to the page information sent by the server side.

For example, in the case of page-level display optimization, the terminal may select a target page template to be used for page rendering among a plurality of page templates according to the indication of the page information, and may perform filling of page content by using information of an object to be recommended.

For example, in the case of element-level display optimization, the terminal may perform dynamic rendering according to the page elements indicated by the page information and the element materials corresponding to the page elements, on the basis of a basic page template supporting free combination of the page elements.

It should be noted that, for specific contents of the page information, reference may be made to the related description of the embodiment shown in fig. 3, and details are not repeated here.

According to the recommendation method provided by the embodiment of the application, the page information sent by the server side is received, the page information is determined by the server side aiming at the page display of the object to be recommended of the user, the page information is used for indicating the style of the page displayed by the object to be recommended, and the page display is carried out on the object to be recommended according to the page information.

Fig. 13 is a schematic structural diagram of a recommendation device according to an embodiment of the present application; referring to fig. 13, the present embodiment provides a recommendation apparatus, which may execute the recommendation method shown in fig. 3, and specifically, the recommendation apparatus may include:

the determining module 131 is configured to determine page information for page display of an object to be recommended of a user, where the page information is used to indicate a style of a page on which the object to be recommended is displayed;

a sending module 132, configured to send the page information to a terminal used by the user, so that the terminal performs page display on the object to be recommended according to the page information.

Optionally, the determining module 131 is specifically configured to select one page template from a plurality of preset page templates as a target page template according to the user characteristics of the user, so as to obtain the page information, where the target page template corresponds to a preset page style, and the page information is specifically used to indicate the target page template.

Optionally, the determining module 131 is specifically configured to select, as a page material to be displayed, a part of candidate materials meeting a certain condition from the multiple candidate materials of the object to be recommended, where the page material corresponds to a page element, and the page element corresponds to a preset element style; and generating the page information based on the page material, wherein the page information is specifically used for indicating the page material and the corresponding page elements thereof.

Optionally, the determining module 131 is configured to select, as a page material to be displayed, a partial candidate material that meets a certain condition from the multiple candidate materials of the object to be recommended, and specifically includes:

Optionally, the determining module 131 is configured to determine a material score of each candidate material based on a feedback condition of the target group for each candidate material of the object to be recommended, and specifically includes:

and respectively inputting the crowd characteristics of the target crowd and the material characteristics of the candidate materials of the object to be recommended into a pre-trained linear model to obtain the material score of each candidate material.

determining a material score of each candidate material based on the feedback condition of the user for each candidate material of the object to be recommended; and according to the material scores of the candidate materials, determining the candidate materials with the scores meeting certain conditions from the candidate materials as the page materials to be displayed.

Optionally, the determining module 131 is configured to determine a material score of each candidate material based on a feedback condition of the user for each candidate material of the object to be recommended, and specifically includes:

and respectively inputting the user characteristics of the user, the object characteristics of the object to be recommended and the material characteristics of each candidate material of the object to be recommended into a pre-trained neural network model to obtain the material score of each candidate material.

Optionally, the neural network model comprises a first sub-network and a second sub-network;

the first sub-network is used for determining pairwise association relations among the user characteristics, the object characteristics and the material characteristics;

and the second sub-network is used for fusing the pairwise associations to obtain a final association relation, and the final association relation is used for representing the material score.

Optionally, the apparatus further includes a first obtaining module, configured to obtain the candidate materials based on a recommendation scene corresponding to the object to be recommended.

Optionally, the apparatus further includes a second obtaining module, configured to obtain the candidate materials by calling an obtaining interface provided by the material service platform.

Optionally, the candidate materials include new materials that are online within a latest preset time length;

the determining module 131 is configured to select a part of candidate materials from the multiple candidate materials of the object to be recommended as page materials to be displayed, and specifically includes: and selecting the new material from the candidate materials of the object to be recommended as a page material to be displayed according to a certain probability.

Optionally, the candidate materials include: the image processing method comprises the steps of obtaining an original image material and a generated image material obtained based on the original image material.

Optionally, the original picture material is uploaded by the provider of the object to be recommended through an object provider platform.

Optionally, the apparatus further comprises: and the processing module is used for carrying out preset processing on the original picture material, displaying the picture obtained after the preset processing to the provider through the object provider platform, and modifying or confirming the picture by the provider through the object provider platform so as to obtain the generated picture material.

Optionally, the apparatus further comprises: and the third acquisition module is used for acquiring user feedback statistics aiming at the original picture material and the generated picture material respectively and prompting the user feedback statistics to the provider through the object provider platform.

Optionally, the candidate material includes a text material; the device further comprises: and the obtaining module is used for obtaining the text material based on the object text of the object to be recommended.

Optionally, the obtaining module is specifically configured to determine, based on a preset named entity tag set, a plurality of previous words with high occurrence rates, of the named entity tags belonging to the named entity tag set, from the object text of the object to be recommended; and extracting short texts containing the words from the long texts of the objects to be recommended.

Optionally, the named entity tag set corresponds to a recommendation scenario.

Optionally, the obtaining module is configured to extract a short text containing the multiple words from the long text of the object to be recommended, and specifically includes: and extracting short texts which contain the words and do not contain the words in a preset negative word list from the long texts of the objects to be recommended.

Optionally, the determining module is configured to select a part of candidate materials from the multiple candidate materials of the object to be recommended as page materials to be displayed, and specifically includes: receiving a trigger message, wherein the trigger message is used for indicating candidate materials corresponding to at least one page element respectively; the candidate elements are at least partial candidate elements of the object to be recommended; and under the trigger of the trigger message, selecting partial candidate materials from the candidate materials of the object to be recommended as page materials to be displayed.

Optionally, the trigger message is further configured to indicate a selection rule of a page element in a page, where meeting a certain condition includes that the corresponding page element meets the requirement of the selection rule.

Optionally, the determining module is configured to generate the page information based on the page material, and specifically includes: determining various material combination modes of the page materials according to the page materials; and selecting a material combination mode from the multiple material combination modes, and generating page information according to the selected material combination mode.

The apparatus shown in fig. 13 can execute the method of the embodiment shown in fig. 3, and reference may be made to the related description of the embodiment shown in fig. 3 for a part of this embodiment that is not described in detail. The implementation process and technical effect of the technical solution refer to the description in the embodiment shown in fig. 3, and are not described herein again.

In one possible implementation, the structure of the recommendation device shown in fig. 13 may be implemented as a computer device. As shown in fig. 14, the computer apparatus may include: a processor 141 and a memory 142. Wherein the memory 142 is used for storing a program for supporting a computer device to execute the recommended method provided in the embodiment shown in fig. 3, and the processor 141 is configured for executing the program stored in the memory 142.

The program comprises one or more computer instructions which, when executed by processor 141, enable the following steps to be performed:

Optionally, processor 141 is further configured to perform all or part of the steps in the foregoing embodiment shown in fig. 3.

The computer device may further include a communication interface 143 for communicating with other devices or a communication network.

Fig. 15 is a schematic structural diagram of a recommendation device according to another embodiment of the present application; referring to fig. 15, the present embodiment provides a recommendation apparatus, which may execute the recommendation method shown in fig. 12, and specifically, the recommendation apparatus may include:

a receiving module 151, configured to receive page information sent by a server side, where the page information is determined by the server side for page display of an object to be recommended of a user, and the page information is used to indicate a style of a page on which the object to be recommended is displayed;

and a display module 152, configured to perform page display on the object to be recommended according to the page information.

The apparatus shown in fig. 15 can execute the method of the embodiment shown in fig. 12, and reference may be made to the related description of the embodiment shown in fig. 12 for a part of this embodiment that is not described in detail. The implementation process and technical effect of the technical solution refer to the description in the embodiment shown in fig. 12, and are not described herein again.

In one possible implementation, the structure of the recommendation device shown in fig. 15 may be implemented as a terminal. As shown in fig. 16, the terminal may include: a processor 161 and a memory 162. Wherein, the memory 162 is used for storing a program for supporting the terminal to execute the recommended method provided in the embodiment shown in fig. 13, and the processor 161 is configured for executing the program stored in the memory 162.

The program comprises one or more computer instructions which, when executed by the processor 161, are capable of performing the steps of:

Optionally, the processor 161 is further configured to perform all or part of the steps in the foregoing embodiment shown in fig. 12.

The terminal may further include a communication interface 163, which is used for the terminal to communicate with other devices or a communication network.

In addition, the present application provides a computer storage medium for storing computer software instructions for a computer device, which includes a program for executing the recommendation method in the method embodiment shown in fig. 3.

The embodiment of the present application provides a computer storage medium for storing computer software instructions for a terminal, which includes a program for executing the recommendation method in the method embodiment shown in fig. 12.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by adding a necessary general hardware platform, and of course, can also be implemented by a combination of hardware and software. With this understanding in mind, the above-described technical solutions and/or portions thereof that contribute to the prior art may be embodied in the form of a computer program product, which may be embodied on one or more computer-usable storage media having computer-usable program code embodied therein (including but not limited to disk storage, CD-ROM, optical storage, etc.).

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

Finally, it should be noted that: the above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A recommendation method, characterized in that the method comprises:

2. The method according to claim 1, wherein the determining page information for the page display of the object to be recommended of the user comprises:

according to the user characteristics of the user, selecting one page template from a plurality of preset page templates as a target page template to obtain the page information, wherein the target page template corresponds to a preset page style, and the page information is specifically used for indicating the target page template.

3. The method according to claim 1, wherein the determining page information for the page display of the object to be recommended of the user comprises:

selecting partial candidate materials meeting certain conditions from the multiple candidate materials of the object to be recommended as page materials to be displayed, wherein the page materials correspond to page elements, and the page elements correspond to preset element styles;

and generating the page information based on the page materials, wherein the page information is specifically used for indicating the page materials and the corresponding page elements.

4. The method according to claim 3, wherein the selecting partial candidate materials meeting a certain condition from a plurality of candidate materials of the object to be recommended as the page materials to be displayed comprises:

determining a material score of each candidate material based on feedback conditions of different users for each candidate material of the object to be recommended;

and according to the material scores of the candidate materials, determining the candidate materials with the scores meeting certain conditions from the candidate materials as the page materials to be displayed.

5. The method according to claim 3, wherein the selecting partial candidate materials meeting a certain condition from a plurality of candidate materials of the object to be recommended as the page materials to be displayed comprises:

determining a material score of each candidate material based on a feedback condition of a target population for each candidate material of the object to be recommended, wherein the target population is a population to which the user belongs;

6. The method of claim 5, wherein determining the material score of each candidate material of the object to be recommended based on the feedback condition of the target population for each candidate material comprises:

7. The method according to claim 3, wherein the selecting partial candidate materials meeting a certain condition from a plurality of candidate materials of the object to be recommended as the page materials to be displayed comprises:

determining a material score of each candidate material based on the feedback condition of the user for each candidate material of the object to be recommended;

8. The method according to claim 7, wherein the determining a material score of each candidate material of the object to be recommended based on the feedback of the user for each candidate material comprises:

9. The method of claim 8, wherein the neural network model comprises a first sub-network and a second sub-network;

10. The method according to any one of claims 3-9, wherein before selecting a part of the candidate materials from the plurality of candidate materials of the object to be recommended as the page materials to be presented, the method further comprises:

and acquiring the candidate materials based on the recommendation scene corresponding to the object to be recommended.

11. The method according to any one of claims 3-9, wherein before selecting a part of the candidate materials from the plurality of candidate materials of the object to be recommended as the page materials to be presented, the method further comprises:

and acquiring the candidate materials by calling an acquisition interface provided by a material service platform.

12. The method according to any one of claims 3-9, wherein the plurality of candidate materials includes new materials that have come online within a last preset time period;

the selecting part of candidate materials from the candidate materials of the object to be recommended as page materials to be displayed comprises:

and selecting the new material from the candidate materials of the object to be recommended as a page material to be displayed according to a certain probability.

13. The method of any of claims 3-9, wherein the candidate materials comprise: the image processing method comprises the steps of obtaining an original image material and a generated image material obtained based on the original image material.

14. The method of claim 13, wherein the raw picture material is uploaded by a provider of the object to be recommended through an object provider platform.

15. The method of claim 14, further comprising:

and presetting the original picture material, and displaying the picture obtained after the presetting to the provider through the object provider platform so that the provider modifies or confirms the picture through the object provider platform to obtain the generated picture material.

16. The method of claim 14, further comprising:

and acquiring user feedback statistics aiming at the original picture material and the generated picture material respectively, and prompting the user feedback statistics to the provider through the object provider platform.

17. The method of any of claims 3-9, wherein the candidate material comprises text material;

the method further comprises the following steps: and obtaining the text material based on the object text of the object to be recommended.

18. The method according to claim 17, wherein the obtaining the text material based on the object text of the object to be recommended comprises:

determining a plurality of front words of which the named entity tags belong to the named entity tag set and the occurrence rate is high from the object text of the object to be recommended based on a preset named entity tag set;

and extracting short texts containing the words from the long texts of the objects to be recommended.

19. The method of claim 18, wherein the set of named entity tags corresponds to a recommendation scenario.

20. The method of claim 19, wherein the extracting a short text containing the plurality of words from the long text of the object to be recommended comprises:

and extracting short texts which contain the words and do not contain the words in a preset negative word list from the long texts of the objects to be recommended.

21. The method according to any one of claims 3-9, wherein the selecting partial candidate materials from a plurality of candidate materials of the object to be recommended as page materials to be presented comprises:

receiving a trigger message, wherein the trigger message is used for indicating candidate materials corresponding to at least one page element respectively; the candidate elements are at least partial candidate elements of the object to be recommended;

and under the trigger of the trigger message, selecting partial candidate materials from the candidate materials of the object to be recommended as page materials to be displayed.

22. The method of claim 21, wherein the trigger message is further used to indicate a selection rule of a page element in a page, and wherein the meeting a certain condition includes the corresponding page element meeting the selection rule requirement.

23. The method of any of claims 3-9, wherein generating the page information based on the page material comprises:

determining various material combination modes of the page materials according to the page materials;

and selecting a material combination mode from the multiple material combination modes, and generating page information according to the selected material combination mode.

24. A recommendation method, characterized in that the method comprises:

25. A recommendation device, comprising:

26. A recommendation device, comprising:

27. A computer device, comprising: a memory, a processor; wherein the memory is to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of any of claims 1 to 23.

28. A terminal, comprising: a memory, a processor; wherein the memory is to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of claim 24.