Propaganda image generation method and device, equipment, medium and product

By receiving the client's promotional image acquisition request, obtaining and matching the user's historical tags and material libraries, personalized promotional images are generated, which solves the problem that promotional images cannot be personalized in the existing technology, and improves the attractiveness of user experience and content display .

CN118964655APending Publication Date: 2024-11-15BEIJING QIYI CENTURY SCI & TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411022624.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The promotional images of existing video resources cannot provide personalized displays for different users, resulting in a low user experience.

Method used

By receiving the client's promotional image acquisition request, obtain the target account historical tags and multimedia content material library of logged in to the client, determine the materials matching the historical tags from the material library, and generate personalized promotional images. The method includes matching the material using a pre-trained recall model, background extraction model, and role recognition model, and obtaining historical labels in combination with the account preference model to generate promotional content for user interests.

Benefits of technology

It realizes the provision of personalized promotional images for each user, improves the user experience and interest in multimedia content, reduces manual processing, and improves efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964655A_ABST
    Figure CN118964655A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a propaganda image generation method and device, equipment, a medium and a product, and relates to the technical field of data processing. The method comprises the following steps: receiving an acquisition request for a propaganda image of multimedia content sent by a client; acquiring a historical label of a target account logging in the client, and acquiring a material library corresponding to the multimedia content; one material library corresponds to one type of layer in the propaganda image, and different material libraries correspond to different types of layers; determining a material matched with the historical label from each material library as a to-be-utilized material corresponding to the material library; and taking each to-be-utilized material as an element of the type corresponding to the material library to which the to-be-utilized material belongs in the propaganda image, and generating the propaganda image of the multimedia content displayed at the client. According to the invention, the user experience can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, equipment, medium and product for generating a promotional image. Background Art

[0002] The client can display promotional images of various video resources, and users can learn about video resources through the promotional images, and then select a video resource to be played from multiple video resources. Among them, the promotional image can be a poster image.

[0003] However, the promotional images of video resources are usually pre-set and will not change. That is, for each video resource, the promotional image displayed by the client to all users is the same, and different promotional images are not provided to each user in a targeted manner, resulting in a poor user experience. Summary of the invention

[0004] The purpose of the embodiment of the present invention is to provide a method for generating a promotional image to improve the user experience. The specific technical solution is as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for generating a promotional image, which is applied to a server. The method includes:

[0006] Receiving a request for obtaining a promotional image of multimedia content sent by a client;

[0007] Obtaining a history tag of a target account that logs into the client, and obtaining a material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements;

[0008] Determine, from each material library, a material matching the historical tag as the to-be-used material corresponding to the material library;

[0009] Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in a promotional image, so as to generate a promotional image for the multimedia content displayed on the client.

[0010] Optionally, determining the material matching the historical tag from each material library as the to-be-used material corresponding to the material library includes:

[0011] For each material library, the material library and the historical label are input into a pre-trained recall model to obtain the similarity between each material in the material library and the historical label output by the recall model; wherein the recall model is trained based on the sample material, the historical label of the first sample account, and the label used to characterize the access result of the first sample account to the sample material;

[0012] The material in the material library with the greatest similarity to the historical label is used as the to-be-used material corresponding to the material library.

[0013] Optionally, the material library includes at least one of the following:

[0014] A material library containing elements of a background type corresponding to the multimedia content and a material library containing elements of a role type corresponding to the multimedia content.

[0015] Optionally, a method for acquiring a material library containing elements of a background type corresponding to the multimedia content includes:

[0016] Inputting a plurality of associated images of the multimedia content into a pre-trained background extraction model respectively, obtaining an element of a background type of each associated image output by the background extraction model, so as to obtain a material library containing elements of the background type corresponding to the multimedia content; wherein the background extraction model is trained based on a first sample image and a label for indicating a position of an element of the background type in the first sample image;

[0017] and / or,

[0018] The method for acquiring a material library containing elements of the role type corresponding to the multimedia content includes:

[0019] Input multiple associated images of the multimedia content into a pre-trained role recognition model to obtain the role type elements of each associated image output by the role recognition model, so as to obtain a material library containing the elements of the role type corresponding to the multimedia content; wherein the role recognition model is trained based on a second sample image and a label for representing the position of the role type element in the second sample image.

[0020] Optionally, obtaining a history tag of a target account that logs into the client includes:

[0021] Obtaining historical behavior data of a target account that logs into the client; wherein the historical behavior data includes at least one of the following: historically watched videos, ratings of videos, comments on videos, and historical search content;

[0022] The historical behavior data is input into a pre-trained account preference model to obtain the historical tags output by the account preference model; wherein the account preference model is trained based on the historical behavior data of the second sample account and the interest tags of the second sample account.

[0023] Optionally, before using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image of the multimedia content for display on the client, the method further includes:

[0024] In the promotional image template library, determine the template that matches each material to be used as the target template;

[0025] The step of using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image for the multimedia content displayed on the client includes:

[0026] Each material to be used is used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and is added to a position preset for the element of the type in the target template to obtain the promotional image of the multimedia content.

[0027] Optionally, determining a template matching each material to be used in the promotional image template library as a target template includes:

[0028] According to the interface display rules of the client, the size of each determined material to be used is adjusted;

[0029] In the promotional image template library, a template matching each resized material to be used is determined as a target template.

[0030] Optionally, the step of using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image of the multimedia content for display on the client includes:

[0031] Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and combined with the name of the multimedia content, a promotional image of the multimedia content for display on the client is generated.

[0032] Optionally, the step of using each to-be-used material as an element of a type corresponding to a material library to which the to-be-used material belongs in the promotional image, and combining the name of the multimedia content to generate a promotional image of the multimedia content for display on the client includes:

[0033] Obtaining a text of the name of the multimedia content in the language format required by the client;

[0034] Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and combined with the acquired text, a promotional image of the multimedia content for display on the client is generated.

[0035] Optionally, there are multiple history tags, and the categories of the history tags include at least one of the following categories: theme preference features for content classifications representing account preferences, role preference features for content roles representing account preferences, and plot preference features for content plots representing account preferences.

[0036] In a second aspect, an embodiment of the present invention provides a promotional image generating device, the device comprising:

[0037] A receiving module, used for receiving a request for obtaining a promotional image of multimedia content sent by a client;

[0038] An acquisition module, used to acquire the history tags of the target account that logs into the client, and to acquire the material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements;

[0039] A determination module, used to determine the material matching the historical tag from each material library as the to-be-used material corresponding to the material library;

[0040] The generating module is used to use each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and to generate a promotional image of the multimedia content for display on the client.

[0041] An embodiment of the present invention further provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0042] Memory, used to store computer programs;

[0043] The processor is used to implement the above-mentioned promotional image generation method when executing the program stored in the memory.

[0044] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned promotional image production method is implemented.

[0045] An embodiment of the present invention further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the computer program implements the above-mentioned promotional image generation method.

[0046] A promotional image generation method provided by an embodiment of the present invention can receive a request for obtaining a promotional image for multimedia content sent by a client, obtain the historical tag of a target account that logs into the client, and obtain a material library corresponding to the multimedia content; determine the material matching the historical tag from each material library as the to-be-used material corresponding to the material library; use each to-be-used material as an element of the type corresponding to the material library to which the to-be-used material belongs in the promotional image, and generate a promotional image of the multimedia content for display on the client.

[0047] Based on the above processing, since one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements, the material to be used determined from each material library corresponds to one type of element in the promotional image; since the material to be used is the material matching the historical tag, each material to be used that matches the historical tag is used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and the material matching the historical tag can be used to generate a promotional image of multimedia content matching the historical tag; since the historical tag of the target account can represent the preference of the target account, promotional content of the multimedia content that the target account is interested in can be generated for the target account, and different promotional images can be provided to each account in a targeted manner to improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0049] Figure 1 It is a flowchart of a first method for generating a promotional image provided in an embodiment of the present invention;

[0050] Figure 2 It is a flowchart of a second method for generating a promotional image provided in an embodiment of the present invention;

[0051] Figure 3 A schematic flow chart of a process of obtaining historical tags of a target account in a method for generating a promotional image provided in an embodiment of the present invention;

[0052] Figure 4 It is a flowchart of a third method for generating a promotional image provided in an embodiment of the present invention;

[0053] Figure 5 It is a flowchart diagram of a fourth method for generating a promotional image provided in an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of the principle of a method for generating a promotional image provided in an embodiment of the present invention;

[0055] Figure 7 It is a structural schematic diagram of a promotional image generating device provided in an embodiment of the present invention;

[0056] Figure 8 The figure is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0058] In order to better understand the embodiments of the present invention, the prior art is introduced below.

[0059] With the rapid development of digital media, users have an increasing demand for personalized video viewing. In order to facilitate users to understand the video resources they are watching, posters of video resources can be displayed to users.

[0060] Generally speaking, posters of video resources are pre-designed and generated by the video resource providers. For each video resource, the content of the posters shown to different users is the same. This poster display method is monotonous and cannot meet users' needs for personalized experience, resulting in a low user experience.

[0061] In order to improve the user experience, the embodiments of the present invention provide a method, device, equipment, medium and product for generating a promotional image.

[0062] The promotional image generation method provided by the embodiment of the present invention may be applied to an electronic device, and the electronic device may be a server for responding to a client request. In a specific application, the electronic device may be a server.

[0063] A method for generating a promotional image provided by an embodiment of the present invention may include the following steps:

[0064] Receiving a request for obtaining a promotional image of multimedia content sent by a client;

[0065] Obtain the history tag of the target account that logs into the client, and obtain the material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements;

[0066] Determine the material matching the historical tag from each material library as the to-be-used material corresponding to the material library;

[0067] Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, so as to generate a promotional image of multimedia content for display on the client.

[0068] Based on the above processing, since one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements, the material to be used determined from each material library corresponds to one type of element in the promotional image; since the material to be used is the material matching the historical tag, each material to be used that matches the historical tag is used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and the material matching the historical tag can be used to generate a promotional image of multimedia content matching the historical tag; since the historical tag of the target account can represent the preference of the target account, promotional content of the multimedia content that the target account is interested in can be generated for the target account, and different promotional images can be provided to each account in a targeted manner to improve the user experience.

[0069] The following describes a method for generating a promotional image provided by an embodiment of the present invention in conjunction with the accompanying drawings. Figure 1 As shown, the method may include steps S101-S104.

[0070] S101: Receive a request for obtaining a promotional image of multimedia content sent by a client.

[0071] It is understandable that when the user logs in to the target account on the client and opens the interface for displaying the promotional image of the multimedia content, the server can receive the acquisition request for the promotional image of the multimedia content sent by the client. The multimedia content can be audio, video, or text and pictures. If the multimedia content is a video, the video can be one of the videos to be played in the display interface of the client; the promotional image can be an image containing the content of the video, for example, the video is a movie, and the promotional image can be a poster containing the actor photos and the title of the movie. If the multimedia content is audio, the audio can be one of the audios to be played in the display interface of the client; the promotional image can be an image showing the audio content, for example, the audio tells a fairy tale, and the promotional image can be a poster containing an image generated according to the plot of the fairy tale and the name of the fairy tale. If the multimedia content is text and pictures, the text and pictures can be one of the text and pictures to be displayed in the display interface of the client; the promotional image can be an image containing the content of the text and pictures, for example, the promotional image can be a poster containing an introduction to the text content of the text and pictures and a picture in the picture content of the text and pictures.

[0072] S102, obtaining the history tags of the target account that logs into the client, and obtaining the material library corresponding to the multimedia content.

[0073] Among them, one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements.

[0074] It is understandable that the user can log in to the client with the target account. When the user opens the interface of the client promotional image, the client sends a request to the server. The request can include the identity of the target account. The server can obtain the historical tags of the target account in the database storing the historical tags of the target account based on the identity of the target account in the received request. In addition, the request is for the promotional image of the multimedia content. The request can also include the identifier of the multimedia content. The server can obtain the material library corresponding to the multimedia content in the database storing the material library of the multimedia content based on the identifier of the multimedia content. Among them, the material library corresponding to the multimedia content contains the materials required to generate the promotional image. The materials can come from the multimedia content and the content related to the multimedia content. For example, if the multimedia content is a movie, the materials in the material library corresponding to the multimedia content can come from the video frames in the movie, the posters of the movie provided by the film company, the stills of the movie, etc.

[0075] It is understandable that, optionally, the types of elements in the promotional image can be divided according to layers. Exemplarily, the types of elements contained in the promotional image can include background type, character type and information type. The background type element can be a background image showing account preferences in multimedia content, and the layer where the background type element is located can also be called a background layer. The character type can be a character image showing account preferences in multimedia content, and the layer where the character type element is located can also be called a character layer; the information type element can be the name of the multimedia content, and the layer where the information type element is located can also be called an information layer. Optionally, the types of elements in the promotional image can be divided according to pictures, texts, and audios. Exemplarily, the types of elements contained in the promotional image can be picture types, text types, or audio types, and pictures, texts, or audios are combined to generate a dynamic promotional image that can play audio. For example, dynamic pictures, texts, and audios can be combined to generate dynamic posters.

[0076] The history tag of the target account indicates the user interest feature of the target account, that is, the history tag may indicate the interest of the target account, for example, the history tag of the target account may be liking actor A or liking romantic background elements, etc. The material library of multimedia content may be a material library composed of content related to the multimedia content, for example, the material library may include background materials of image frames in the multimedia content, or the material library may include character materials of image frames in the multimedia content.

[0077] It should be noted that there are multiple material libraries, and there can be multiple types of elements in the promotional image. Different types of elements correspond to different material libraries. One material library can correspond to one type of element in the promotional image, that is, the materials in one material library can be used as elements of the type corresponding to the material library in the promotional image. Exemplarily, one material library can be a material library containing only background-type elements of multimedia content, and the material library corresponds to the background-type elements in the promotional image; another material library can be a material library containing only character-type elements of multimedia content, and the material library corresponds to the character-type elements in the promotional image.

[0078] S103: Determine the material matching the historical tag from each material library as the to-be-used material corresponding to the material library.

[0079] It is understandable that for each material library, materials matching the interests of the target account can be determined in the material library according to the interests of the target account indicated by the historical tags of the target account, as the materials to be used corresponding to the material library. For example, if the historical tags indicate that actor A is liked, a character image containing actor A can be determined in the material library containing elements of the role type as the materials to be used corresponding to the material library.

[0080] S104, using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, to generate a promotional image of multimedia content for display on the client.

[0081] It is understandable that by using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, a promotional image containing multiple types of elements can be generated. Furthermore, the generated promotional image can be used for display on the client so that the user can understand the multimedia content through the promotional image.

[0082] Optionally, in one implementation, each material to be used is used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and combined with the name of the multimedia content, a promotional image of the multimedia content for display on the client is generated.

[0083] It is understandable that, in the process of generating a promotional image, the material to be used may be combined with the name of the multimedia content.

[0084] Regarding the process of combining multimedia content to generate a promotional image, in one implementation, after the server determines each material to be used, the server can combine each material to be used with the name of the multimedia content to generate a promotional image of the multimedia content; then the server can send the generated promotional image of the multimedia content to the client to respond to the client's request to obtain the promotional image. Optionally, in another implementation, after the server determines each material to be used, the server can send the determined materials to be used and the name of the multimedia content to the client, and the client combines the received materials to be used and the name of the multimedia content to generate a promotional image of the multimedia content.

[0085] It can be understood that one material library corresponds to one type of element in the promotional image, and one material to be used corresponds to one material library, that is, one material to be used corresponds to one type of element in the promotional image. In addition, the name of the multimedia content also has a layer of its corresponding promotional image. Each material to be used can be used as an element of the type of promotional image it corresponds to, and the name of the multimedia content can be added to the information layer of the promotional image to generate a promotional image of the multimedia content. The name of the multimedia content can be a name used to promote the multimedia content. For example, if the multimedia content is a movie, the name can be the title of the movie.

[0086] Optionally, in one implementation, the process of using each material to be used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and combining it with the name of the multimedia content to generate a promotional image of the multimedia content for display on the client may include steps A1-A2.

[0087] A1, obtaining the text of the multimedia content name in the language format required by the client.

[0088] A2: taking each material to be used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and combining it with the acquired text, generating a promotional image of multimedia content for display on the client.

[0089] It is understandable that the name of the multimedia content in the language format required by the client can be obtained. Exemplarily, the name of the multimedia content in the language format required by the client can be directly obtained. Alternatively, the name of the original multimedia content can be obtained first, and then the name of the multimedia content can be adjusted to match the interface language of the client. For example, if the name of the original multimedia content is "title" in English, and the interface language of the client is Chinese, the name of the multimedia content can be adjusted to "title" in Chinese.

[0090] In order to intuitively display the information of multimedia content to users, the name of the multimedia content can be set above the element of the role type and the element of the background type to ensure that the name of the multimedia content is not blocked, and the element of the role type is set above the element of the background type to ensure that the element of the role type is not blocked by the element of the background type. If there are three layers in the promotional image, namely, the background layer, the role layer and the information layer, the elements of the background type are in the background layer, the elements of the role type are in the role layer, and the name of the multimedia content is in the information layer, the layers to which the materials to be used and the name of the multimedia content are added can be superimposed to obtain a promotional image. Specifically, in order to intuitively display the information of the promotional image to users, the information layer can be set at the top of the promotional image, the role layer can be set in the middle of the promotional image, and the background layer can be set at the bottom of the promotional image. The three layers can be superimposed to obtain the desired promotional image.

[0091] In the embodiment of the present invention, by designing the promotional images in layers, the information of the multimedia content is clearly transmitted to the user, the user's interest in the promotional images is increased, and the click rate of the user on the multimedia content can be increased. In addition, the promotional images can be automatically generated in the embodiment of the present invention, reducing the manual processing process and improving efficiency.

[0092] Based on the above processing, since one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements, the material to be used determined from each material library corresponds to one type of element in the promotional image; since the material to be used is the material matching the historical tag, each material to be used that matches the historical tag is used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and the material matching the historical tag can be used to generate a promotional image of multimedia content matching the historical tag; since the historical tag of the target account can represent the preference of the target account, promotional content of the multimedia content that the target account is interested in can be generated for the target account, and different promotional images can be provided to each account in a targeted manner to improve the user experience.

[0093] Optionally, in another embodiment, Figure 1 Based on the promotional content generation method shown, Figure 2 As shown, step S103 includes steps S1031-S1032.

[0094] S1031, for each material library, the material library and the historical label are input into a pre-trained recall model to obtain the similarity between each material in the material library and the historical label output by the recall model.

[0095] The recall model is obtained by training based on the sample material, the historical labels of the first sample account, and the labels used to characterize the access results of the first sample account to the sample material.

[0096] It is understandable that the recall model can be a model constructed based on a neural network. Exemplarily, the recall model can be a model constructed based on a recurrent neural network (RNN) or a deep neural network (DNN). This embodiment of the present invention is only used as an example and is not specifically limited.

[0097] For a sample material, if the first sample account accesses the sample material, the label used to represent the access result of the first sample account to the sample material may represent access, and exemplarily, the label may be represented by 1; if the first sample account does not access the sample material, the label used to represent the access result of the first sample account to the sample material may represent non-access, and exemplarily, the label may be represented by 0.

[0098] In the process of training the recall model, multiple sample materials and multiple historical labels of the first sample account can be input into the recall model to be trained, and the recall model outputs the similarity between each sample material and each historical label. The label used to indicate that the access result of the first sample account to the sample material is accessed is taken as the true value, and the loss value of the true value and the similarity is calculated. The parameters of the recall model are adjusted based on the loss value, and the recall model is retrained until the recall model reaches the preset convergence condition, thereby obtaining a trained recall model.

[0099] In the process of training the recall model, the model parameters are adjusted to make the similarity output by the recall model as close as possible to the label of the access result characterizing the first sample account accessing the sample material. In other words, the similarity output by the trained recall model can represent the matching degree between the content and material that the target account is interested in, that is, the matching degree between the historical label and the material. For each material library corresponding to the multimedia content, all the materials in the material library and all the historical labels of the target account can be input into the pre-trained recall model to obtain the similarity between each material in the material library and each historical label. For example, for a movie, all the materials 1 and 2 in the material library containing the background type of the movie and the historical labels 1 and 2 of the target account are input into the recall model at the same time, and the similarity between the material 1 and the historical label 1, the similarity between the material 1 and the historical label 2, the similarity between the material 2 and the historical label 1, and the similarity between the material 2 and the historical label 2 output by the recall model are obtained.

[0100] S1032: The material in the material library having the greatest similarity to the historical tag is used as the to-be-used material corresponding to the material library.

[0101] It is understandable that there is a similarity between the historical tag and each material in the material library, and the material with the greatest similarity to the historical tag is the material in the material library that the account is most likely to be interested in. The material in the material library with the greatest similarity to the historical tag can be used as the material to be used corresponding to the material library.

[0102] For example, when Account A logs in to the client, the client homepage needs to display promotional images of 8 TV series. The server can receive a request to obtain promotional images for these 8 TV series. For each TV series, obtain the material library containing background-type materials, the material library containing character-type materials, and the title of the TV series. In addition, obtain all historical tags of Account A. Input all materials in the material library containing background-type materials of the TV series and all historical tags of Account A into the trained recall model, obtain the similarity between each material and each historical tag, and select the material with the greatest similarity as the material to be used. Similarly, input all materials in the material library containing character-type materials of the TV series and all historical tags of Account A into the trained recall model, obtain the similarity between each material and each historical tag, and select the material with the greatest similarity as the material to be used. Combining the background-type material to be used, the character-type material to be used, and the title can generate the promotional image of the TV series. By analogy, the promotional images of these 8 TV series can be generated.

[0103] In the embodiment of the present invention, the material to be used is the material that the user is most likely to be interested in, and the promotional image subsequently generated by using the material to be used can meet the user's interests and improve the user's experience.

[0104] Optionally, in another embodiment, the plurality of material libraries include at least one of the following:

[0105] A material library containing background-type elements corresponding to the multimedia content, and a material library containing role-type elements corresponding to the multimedia content.

[0106] It is understandable that the multiple material libraries may be pre-established material libraries. Exemplarily, a material library containing background-type materials of multimedia content and a material library containing role-type materials of multimedia content may be pre-established respectively.

[0107] The material library containing background-type materials of multimedia content may include background-type materials of associated images of multimedia content. The material library containing character-type materials of multimedia content may include character-type materials in associated images of multimedia content. The associated images of multimedia content include at least one of the following: stills, posters, and image frames in multimedia content, etc.; the characters in the associated images may be actors, actresses, protagonists, or supporting roles, etc.

[0108] The method of obtaining the material library of the background type elements corresponding to the multimedia content includes:

[0109] Multiple associated images of multimedia content are respectively input into a pre-trained background extraction model to obtain background type elements of each associated image output by the background extraction model, so as to obtain a material library containing background type elements corresponding to the multimedia content.

[0110] The background extraction model is obtained by training based on the first sample image and labels for indicating positions of elements of the background type in the first sample image.

[0111] It is understandable that the background extraction model is a model constructed based on a neural network. For example, the background extraction model can be a model constructed based on a convolutional neural network (CNN). In this regard, the embodiments of the present invention are only used as examples and are not specifically limited.

[0112] In the process of training the background extraction model, the first sample image can be input into the background extraction model to be trained to obtain the predicted area of ​​the background type material of the first sample image output by the background extraction model. The loss value is calculated based on the area represented by the label used to represent the position of the background type material in the first sample image and the predicted area of ​​the background type material. The model parameters of the background extraction model are adjusted based on the loss value, and the background extraction model is trained until the background extraction model reaches the preset convergence condition of the model.

[0113] The first sample image and the label used to indicate the position of the background type material in the sample image may be data in an open source annotation dataset, or data in an annotation dataset obtained by manually annotating the drama.

[0114] During the background extraction model training process, the label used to represent the position of the background type material in the first sample image is taken as the true value, and the model parameters are adjusted to make the predicted area as close as possible to the area of ​​the actual background type material in the sample image. Therefore, by inputting multiple associated images into the trained background extraction model, the background type material of each associated image can be extracted.

[0115] The method of obtaining the material library of elements of the role type corresponding to the multimedia content includes:

[0116] Multiple associated images of multimedia content are input into a pre-trained role recognition model to obtain the role type element of each associated image output by the role recognition model, so as to obtain a material library containing the role type elements corresponding to the multimedia content.

[0117] The role recognition model is trained based on the second sample image and labels for indicating positions of elements of role types in the second sample image.

[0118] It is understandable that the role recognition model is a model constructed based on a neural network, and illustratively, the role recognition model may be a model constructed based on a convolutional neural network. In this regard, the embodiment of the present invention is only used as an example and is not specifically limited.

[0119] In the process of training the role recognition model, the second sample image can be input into the role recognition model to be trained to obtain the predicted area of ​​the role type material in the second sample image output by the role recognition model. The loss value is calculated based on the real area represented by the label used to represent the position of the role type material in the second sample image and the predicted area of ​​the role type material. The model parameters of the role recognition model are adjusted based on the loss value, and the role recognition model is trained until the role recognition model reaches the preset convergence condition of the model, and the training of the role recognition model is completed.

[0120] Since in the character recognition model training process, the label used to represent the position of the character type material in the second sample image is taken as the true value, the model parameters are adjusted to make the predicted area of ​​the character type material as close as possible to the actual area of ​​the character type material in the second sample image. Therefore, by inputting multiple related images into the trained character recognition model, the character type material of each related image can be extracted.

[0121] Optionally, if the content of the extracted background type or character type material is incomplete, the content of the background type or character type material can be filled based on a pre-trained image restoration model to obtain a background type or character type material containing complete content.

[0122] It is understandable that the image restoration model may be a neural network model, and illustratively, the image restoration model may be a convolutional neural network model, or a model constructed based on an attention mechanism. In this regard, the embodiments of the present invention are only used as examples and are not specifically limited.

[0123] During the training process of the image restoration model, the third sample image can be input into the image restoration model to be trained to obtain a predicted image output by the image restoration model. A loss value is calculated based on the predicted image and the true value image corresponding to the third sample image. Based on the loss value, the model parameters of the image restoration model are adjusted until the preset convergence conditions of the model are reached, thereby obtaining a trained image restoration model.

[0124] Since in the image restoration model training process, the true value image corresponding to the third sample image is taken as the true value, the model parameters are adjusted to make the predicted image as close to the real image as possible. Therefore, by inputting the background type or character type material into the trained image restoration model, the background type or character type material can be completely restored.

[0125] For example, the content of a background-type material is a tree, but the tree in the material is incomplete. The tree in the material can be filled based on the image restoration model to obtain a material containing a complete tree; or the content of a character-type material is a person, but the person in the material is incomplete. The person in the material can be filled based on the image restoration model to obtain a material containing a complete person.

[0126] In the embodiment of the present invention, a material library of background type and character type materials of multimedia content may be pre-built, so that the background type and character type materials of multimedia content may be subsequently obtained from different material libraries.

[0127] Optionally, in another embodiment, Figure 1 In the promotional image generation method shown, Figure 3 As shown, in step S102, the process of obtaining the historical tags of the target account logged into the client includes steps S1021-S1022.

[0128] S1021, obtaining historical behavior data of the target account that logs into the client.

[0129] The historical behavior data includes at least one of the following: historically watched videos, video ratings, video comments, and historical search content.

[0130] It is understandable that the historical behavior data of the target account logged into the client can be obtained from the database storing the historical behavior data of the target account according to the historical identification of the target account. Among them, the historically watched videos can be the videos watched by the target account within the historical specified time, for example, the historically watched videos can be the videos watched within the past week. The rating of the video can be the rating of the video by the target account, and the comment on the video can be the comment on the video by the target account. The historical search content can be the content searched by the target account within the historical specified time, for example, the title of the play and the name of the person searched by the target account in the past week.

[0131] Optionally, the historically viewed videos may be analyzed to obtain relevant information such as the subject, plot, and actors involved in the historically viewed videos, and the obtained relevant information may be used as part of the historical behavior data.

[0132] S1022, input the historical behavior data into a pre-trained account preference model to obtain the historical labels output by the account preference model.

[0133] The account preference model is trained based on the historical behavior data of the second sample account and the interest tags of the second sample account.

[0134] It can be understood that the account preference model is a model constructed based on a neural network. Exemplarily, the account preference model can be a back propagation (BP) neural network model. In this regard, the embodiments of the present invention are only used as examples and are not specifically limited.

[0135] In the process of training the account preference model, the historical behavior data of the second sample account can be input into the account preference model to be trained, and the account preference model can output the similarity between the account interest and the historical label. According to the interest label and similarity of the second sample account, the loss value is calculated. Based on the loss value, the model parameters of the account preference model are adjusted, and the account preference model is trained until the account preference model reaches the preset convergence condition of the model, thereby obtaining a trained account preference model.

[0136] During the account preference model training process, the interest tag of the second sample account is used as the true value and the model parameters are adjusted to make the account interest output by the prediction model as close as possible to the interest tag of the sample account. Therefore, the historical behavior data is input into the trained account preference model, and the account preference model can output the historical tag.

[0137] Optionally, in one implementation, the number of history tags is multiple, and the categories of history tags include at least one of the following categories: theme preference features for content classifications representing account preferences, role preference features for content roles representing account preferences, and plot preference features for content plots representing account preferences.

[0138] It is understandable that the acquired history tags may be multiple, and multiple interests of the target account may be represented by multiple history tags. Moreover, the history tags may be of different types to represent different types of interests of the target account.

[0139] The categories of history tags may include: theme preference features for multimedia content classifications representing account preferences, role preference features for multimedia content roles representing account preferences, and plot preference features for multimedia content representing account preferences.

[0140] The theme preference feature can be used to indicate the multimedia content classification preferred by the account, for example, the theme preference feature can be the account's preference for suspense or romance. The character preference feature can indicate the account's preference for characters, for example, the character preference feature can indicate the account's preference for actor A or actor B. The plot preference feature can indicate the user's preference for multimedia content, for example, the plot preference feature can indicate the account's preference for love plots or battle plots, etc. In the embodiment of the present invention, multiple categories of historical tags can be obtained, which can improve the accuracy of historical tags.

[0141] In an embodiment of the present invention, historical behavior data is input into a pre-trained account preference model to obtain accurate historical tags, thereby improving the accuracy of subsequently determined materials that match the historical tags and improving the pertinence of the generated promotional images.

[0142] Optionally, in another embodiment, Figure 1 Based on the promotional image generation method shown, Figure 4 As shown, the method further includes step S401, and step S104 includes step S1041.

[0143] S401, in the promotional image template library, determine a template that matches each material to be used as a target template.

[0144] It is understandable that multiple templates adapted to the sizes and contents of different materials can be pre-established to construct a promotional image template library, wherein the size of the constructed template is larger than the size of the material to be placed therein.

[0145] For example, in the process of establishing a promotional image template library, multiple templates can be constructed to which different numbers of elements can be added in the character layer. For example, three elements can be added in the character layer of one constructed template, and two elements can be added in the character layer of another constructed template.

[0146] Accordingly, in the process of selecting a target template, if there are a specified number of materials of the role type that is determined to have the greatest similarity to the historical tag, then a template that can add the specified number of elements in the role layer can be selected, that is, the number of elements in the role layer of the selected template is the same as the number of materials of the determined role type. For example, if there are three materials of the determined role type, then a template that can add three elements in the role layer can be selected as the target template.

[0147] Optionally, in the process of establishing the promotional image template library, multiple templates may be constructed in which the elements in the character layer contain different numbers of characters. For example, the elements in the character layer of one constructed template may contain one character, and the elements in the character layer of another constructed template may contain three characters.

[0148] Accordingly, in the process of selecting a target template, if there are multiple roles in the material of the role layer with the greatest similarity to the historical tag, a template whose elements in the role layer contain multiple roles can be selected as the target template. For example, if there are three roles in the material of the determined role type, a template whose elements in the role layer contain three roles can be selected as the target template.

[0149] Optionally, in the process of establishing a promotional image template library, multiple templates can be constructed in which different numbers of elements can be added to the role layer, and each element contains different numbers of roles. For example, the constructed templates can be: the role layer can add one element, and one element contains a template of one role; or the role layer can add one element, and one element contains templates of three roles; or the role layer can add three elements, and each element contains a template of one role; or the role layer can add three elements, and one element contains one role, and the other elements each contain templates of two roles.

[0150] In the embodiment of the present invention, multiple templates matching the elements in the role layer and the roles contained in the elements can be established to improve the effectiveness of the promotional images generated by the target templates.

[0151] Optionally, in one implementation, behavior recognition can be performed for the characters in the character type material. Exemplarily, the character type material of each associated image can be input into the behavior recognition model to obtain the behavior of the character in the material output by the behavior recognition model. The behavior of the character can be kissing, conflicting, hugging, or back-to-back, etc.

[0152] Among them, the behavior recognition model can be a deep learning model. Exemplarily, the behavior recognition model can be a model built based on a convolutional neural network or a model built based on a recursive neural network, etc. This embodiment of the present invention does not make any specific limitations on this and only provides examples for illustration.

[0153] In the process of training the behavior recognition model, the fourth sample image can be input into the behavior recognition model to be trained to obtain the behavior recognition result output by the behavior recognition model. The loss value is calculated based on the behavior recognition result and the label of the real behavior. The model parameters of the behavior recognition model are adjusted based on the loss value, and the behavior recognition model is trained until the preset convergence conditions of the model are reached to obtain a trained behavior recognition model. Due to the process of training the behavior recognition model, the label of the real behavior is used as the true value, and the model parameters are adjusted so that the behavior recognition result is as close as possible to the behavior represented by the label of the real behavior. Therefore, by inputting the material of the role type into the trained behavior recognition model, the behavior of the role in the material output by the behavior recognition model can be obtained.

[0154] It is understandable that in the process of constructing a template, different layouts of elements corresponding to different behaviors of the characters can be used to construct templates for the layout of materials corresponding to different behaviors. Specifically, the positions of various elements of the character type of the template are arranged according to the positions of different behaviors to obtain templates corresponding to different behaviors. For example, if the behavior of the character is conflicting, the two conflicting characters can be arranged on both sides of the template to generate a template corresponding to the conflicting behavior.

[0155] Accordingly, in the process of selecting the target template, the material of the role type with the greatest similarity to the historical tag can be subjected to behavior recognition to obtain a behavior recognition result. The target template is selected according to the behavior recognition result. For example, if the behavior recognition result is conflict, the template corresponding to the conflicting behavior can be selected.

[0156] Optionally, in the process of selecting a target template, the target template may be selected by combining the behavior recognition result, the number of materials of the determined role type, and the number of roles included in the materials.

[0157] In the embodiment of the present invention, the target template can be accurately selected according to the number of materials, the number of characters in the materials, and the behaviors of the characters in the materials, thereby improving the effectiveness of the promotional image generated by the target template.

[0158] S1041, adding each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to a position preset for the element of the type in the target template to obtain a promotional image of the multimedia content.

[0159] It is understandable that after determining the target template, the position of each element in the target template is fixed, and each material to be used is directly added as an element to the position preset for the element of this type in the target template to obtain a promotional image of the multimedia content.

[0160] In the embodiment of the present invention, a template can be used to generate a promotional image, thereby improving the efficiency of generating the promotional image, reducing the randomness of generating the promotional image, and improving the effectiveness of the generated promotional image.

[0161] Optionally, in another embodiment, Figure 4 Based on the promotional content generation method shown, Figure 5 As shown, step S401 includes steps S4011-S4012.

[0162] S4011, adjusting the size of each determined material to be used according to the interface display rule of the client.

[0163] It is understandable that interface display rules for different clients may be preset; wherein the different clients may be web clients or application clients, etc.

[0164] The interface display rule may be matched with the interface size of the client, and according to the interface display rule, the names of the determined materials to be used and the multimedia content may be adjusted to match the interface size of the client. For example, for a web client, the size of the promotional image to be displayed may be adjusted to match the size of the promotional image display area in the web client; for an application client, the size of the promotional image to be displayed may be adjusted to match the size of the promotional image display area in the application client.

[0165] There are a carousel display area and a general display area in the interface of the web client. The size of the promotional image to be displayed can be directly set to the size of the carousel display area, and the size of the promotional image to be displayed can be set to the standard size of each promotional image arranged in the general display area.

[0166] S4012: In the promotional image template library, a template matching each resized material to be used is determined as a target template.

[0167] It is understandable that after the adjustment of each material to be used is completed, a template matching the material to be used after the size adjustment can be selected as the target template.

[0168] In the embodiment of the present invention, according to the interface display rules of the client, for the promotional image generated after the material is resized, a target template is selected to generate the promotional image, and the client of the target account can be adapted to display the promotional image. In other words, the promotional image generation method provided in the embodiment of the present invention can generate personalized promotional images that the target account likes for distribution in various scenarios such as different languages ​​and different clients, increase the stickiness between users and video content, improve user experience, and thus improve user satisfaction and retention rate.

[0169] In order to more clearly understand the embodiments of the present invention, Figure 6 A method for generating a promotional image provided by an embodiment of the present invention is introduced.

[0170] like Figure 6 As shown, the promotional image of the multimedia content can be a poster. The source of the poster material can include at least one of the following: stills, posters provided by the film company, and image frames in the multimedia content, that is, the source of the material includes: film company, video frames, stills, etc.

[0171] In order to realize poster generation, the poster can be layered into a background layer, a character layer and an information layer.

[0172] In order to obtain a material library containing background type materials of posters, multiple associated images of multimedia content can be input into a pre-trained background extraction model respectively to obtain background type materials of each associated image output by the background extraction model, so as to obtain a material library containing background image type materials of multimedia content.

[0173] In order to obtain a material library containing character type materials of posters, multiple associated images of multimedia content can be input into a pre-trained character recognition model to obtain character type materials of each associated image output by the character recognition model to obtain a material library containing character type materials of multimedia content.

[0174] It is understandable that the background type material of the poster and the character type material of the poster can be intelligently cropped. Specifically, the material can be intelligently cropped according to the interface size of the poster displayed in the client. In addition, the background type material of the poster and the character type material of the poster can be labeled. For example, the background type material can be labeled and identified as a suspense type material; the character type material can be labeled and identified as the material of actor A. In the label recognition process, labels can be set in advance according to the preference characteristics of the target account, and label recognition can be performed for the background type and character type materials in the poster.

[0175] Optionally, a material library containing character type materials of the poster and a material library containing background type materials of the poster can be stored in a candidate server, which can store materials used to generate the poster. The candidate server is also called a poster layered candidate pool. In an embodiment of the present invention, posters can be generated for overseas clients through the materials in the candidate server.

[0176] In order to obtain the historical tags of the target account, the historical behavior data of the target account logged into the client can be obtained; specifically, the historical behavior collection process can be performed in advance to obtain the historical viewing videos, video ratings, video comments and historical search content. Then, the historical behavior data is input into the pre-trained account preference model to obtain the historical tags output by the account preference model. Among them, the account preference model includes an account feature extraction module and a preference modeling module. The account feature extraction module can extract theme preference features, character preference features and plot preference features; the preference modeling module is a module set based on clustering algorithms, classification algorithms and regression algorithms. Among them, historical behavior collection can be to collect historical behaviors from the database of historical behaviors of the target account.

[0177] In the process of poster recall, for each material library, the material library and the historical label can be input into the pre-trained recall model to obtain the similarity between each material in the material library and the historical label output by the recall model; the material with the greatest similarity to the historical label in the material library is used as the material to be used corresponding to the material library. That is, the material to be used of the character type of the poster and the material to be used of the background type of the poster are obtained. In the process of poster recall, the material in the material library can be obtained from the poster layered candidate pool, and the historical label can be obtained from the preference modeling module. Then, the material in the material library and the historical label are input into the recall model to determine the material to be used.

[0178] In the process of combining and sorting posters, the server can combine the names of the determined materials to be used and the multimedia content to generate a poster of the multimedia content for displaying to the user. Optionally, the server can send the determined materials to be used to the client, and the client combines the names of the determined materials to be used and the multimedia content to generate a poster. In the process of combining and sorting posters, the historical tags can be obtained from the preference modeling module to combine and sort each poster.

[0179] Based on the above processing, since one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements, the material to be used determined from each material library corresponds to one type of element in the promotional image; since the material to be used is the material matching the historical tag, each material to be used that matches the historical tag is used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and the material matching the historical tag can be used to generate a promotional image of multimedia content matching the historical tag; since the historical tag of the target account can represent the preference of the target account, promotional content of the multimedia content that the target account is interested in can be generated for the target account, and different promotional images can be provided to each account in a targeted manner to improve the user experience.

[0180] The embodiment of the present invention also provides a promotional image generating device, such as Figure 7 As shown, the device comprises:

[0181] The receiving module 710 is used to receive a request for obtaining a promotional image of multimedia content sent by a client;

[0182] The acquisition module 720 is used to acquire the history tags of the target account that logs into the client, and acquire the material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements;

[0183] A determination module 730 is used to determine a material matching the historical tag from each material library as the material to be used corresponding to the material library;

[0184] The generating module 740 is used to use each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and to generate a promotional image of multimedia content for display on the client.

[0185] Optionally, determine the modules, including:

[0186] An input unit, for each material library, inputting the material library and the historical label into a pre-trained recall model, and obtaining the similarity between each material in the material library and the historical label output by the recall model; wherein the recall model is trained based on the sample material, the historical label of the first sample account, and the label used to characterize the access result of the first sample account to the sample material;

[0187] The first determining unit is used to use the material in the material library that has the greatest similarity to the historical tag as the to-be-used material corresponding to the material library.

[0188] Optionally, the multiple material libraries include at least one of the following:

[0189] A material library containing background-type elements corresponding to the multimedia content, and a material library containing role-type elements corresponding to the multimedia content.

[0190] Optionally, a method for acquiring a material library of background-type elements corresponding to multimedia content includes:

[0191] Inputting multiple associated images of multimedia content into a pre-trained background extraction model respectively, obtaining an element of the background type of each associated image output by the background extraction model, so as to obtain a material library containing elements of the background type corresponding to the multimedia content; wherein the background extraction model is trained based on the first sample image and a label for indicating the position of the element of the background type in the first sample image;

[0192] and / or,

[0193] Methods for obtaining a material library containing elements of role types corresponding to multimedia content include:

[0194] Input multiple associated images of multimedia content into a pre-trained role recognition model to obtain the role type elements of each associated image output by the role recognition model, so as to obtain a material library containing elements of the role type corresponding to the multimedia content; wherein the role recognition model is trained based on a second sample image and a label for indicating the position of the role type element in the second sample image.

[0195] Optionally, obtain the history tags of the target account that logged into the client, including:

[0196] Obtain historical behavior data of the target account logged into the client; wherein the historical behavior data includes at least one of the following: historically watched videos, ratings of videos, comments on videos, and historical search content;

[0197] The historical behavior data is input into a pre-trained account preference model to obtain the historical labels output by the account preference model; wherein the account preference model is trained based on the historical behavior data of the second sample account and the interest labels of the second sample account.

[0198] Optionally, the device further comprises:

[0199] The second determining unit is used to determine, in the promotional image template library, a template that matches each material to be used as a target template.

[0200] Generate modules, including:

[0201] The adding unit is used to add each material to be used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image to a position preset for the element of the type in the target template to obtain a promotional image of the multimedia content.

[0202] The second determination module is specifically used to adjust the size of each determined material to be used according to the interface display rule of the client;

[0203] In the promotional image template library, a template matching each resized material to be used is determined as a target template.

[0204] The generated modules include:

[0205] The combining unit is used to use each material to be used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and to combine it with the name of the multimedia content to generate a promotional image of the multimedia content for display on the client.

[0206] The combining unit is specifically used to obtain the text of the name of the multimedia content in the language format required by the client;

[0207] Each material to be used is used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and combined with the acquired text, a promotional image of multimedia content for display on the client is generated.

[0208] Optionally, there are multiple history tags, and the categories of history tags include at least one of the following categories: theme preference features for content classifications representing account preferences, role preference features for content roles representing account preferences, and plot preference features for content plots representing account preferences.

[0209] Based on the above processing, since one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements, the material to be used determined from each material library corresponds to one type of element in the promotional image; since the material to be used is the material matching the historical tag, each material to be used that matches the historical tag is used as an element of the type corresponding to the material library to which the material to be used belongs in the promotional image, and the material matching the historical tag can be used to generate a promotional image of multimedia content matching the historical tag; since the historical tag of the target account can represent the preference of the target account, promotional content of the multimedia content that the target account is interested in can be generated for the target account, and different promotional images can be provided to each account in a targeted manner to improve the user experience.

[0210] The embodiment of the present invention further provides an electronic device, such as Figure 8 As shown, it includes a processor 801, a communication interface 802, a memory xx3 and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other through the communication bus 804.

[0211] Memory 803, used for storing computer programs;

[0212] The processor 801 is used to implement the above-mentioned promotional image generation method when executing the program stored in the memory 803.

[0213] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0214] The communication interface is used for communication between the above terminal and other devices.

[0215] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0216] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0217] In another embodiment of the present invention, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the promotional image generating method described in any one of the above embodiments is implemented.

[0218] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is run on a computer, the computer executes the method for generating a promotional image described in any one of the above embodiments.

[0219] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0220] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0221] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0222] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for generating a promotional image, characterized in that: Applied to a server, the method comprises: Receiving a request for obtaining a promotional image of multimedia content sent by a client; Obtaining a history tag of a target account that logs into the client, and obtaining a material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements; Determine, from each material library, a material matching the historical tag as the to-be-used material corresponding to the material library; Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in a promotional image, so as to generate a promotional image for the multimedia content displayed on the client.

2. The method according to claim 1, characterized in that The step of determining the material matching the historical tag from each material library as the to-be-used material corresponding to the material library includes: For each material library, the material library and the historical label are input into a pre-trained recall model to obtain the similarity between each material in the material library and the historical label output by the recall model; wherein the recall model is trained based on the sample material, the historical label of the first sample account, and the label used to characterize the access result of the first sample account to the sample material; The material in the material library with the greatest similarity to the historical label is used as the to-be-used material corresponding to the material library.

3. The method according to claim 2, characterized in that The material library includes at least one of the following: A material library containing elements of a background type corresponding to the multimedia content and a material library containing elements of a role type corresponding to the multimedia content.

4. The method according to claim 3, characterized in that The method for acquiring a material library containing elements of a background type corresponding to the multimedia content includes: Inputting a plurality of associated images of the multimedia content into a pre-trained background extraction model respectively, obtaining an element of a background type of each associated image output by the background extraction model, so as to obtain a material library containing elements of the background type corresponding to the multimedia content; wherein the background extraction model is trained based on a first sample image and a label for indicating a position of an element of the background type in the first sample image; and / or, The method for acquiring a material library containing elements of the role type corresponding to the multimedia content includes: Input multiple associated images of the multimedia content into a pre-trained role recognition model to obtain the role type elements of each associated image output by the role recognition model, so as to obtain a material library containing the elements of the role type corresponding to the multimedia content; wherein the role recognition model is trained based on a second sample image and a label for representing the position of the role type element in the second sample image.

5. The method according to claim 1, characterized in that The step of obtaining the history tags of the target account that logs into the client includes: Obtaining historical behavior data of a target account that logs into the client; wherein the historical behavior data includes at least one of the following: historically watched videos, ratings of videos, comments on videos, and historical search content; The historical behavior data is input into a pre-trained account preference model to obtain the historical tags output by the account preference model; wherein the account preference model is trained based on the historical behavior data of the second sample account and the interest tags of the second sample account.

6. The method according to claim 1, characterized in that Before using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image of the multimedia content for display on the client, the method further includes: In the promotional image template library, determine the template that matches each material to be used as the target template; The step of using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image for the multimedia content displayed on the client includes: Each material to be used is used as an element of a type corresponding to the material library to which the material to be used belongs in the promotional image, and is added to a position preset for the element of the type in the target template to obtain the promotional image of the multimedia content.

7. The method according to claim 6, characterized in that The step of determining, in the promotional image template library, a template matching each material to be used as a target template includes: According to the interface display rules of the client, the size of each determined material to be used is adjusted; In the promotional image template library, a template matching each resized material to be used is determined as a target template.

8. The method according to claim 1, characterized in that: The step of using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image to generate a promotional image for the multimedia content displayed on the client includes: Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and combined with the name of the multimedia content, a promotional image of the multimedia content for display on the client is generated.

9. The method according to claim 8, characterized in that The method of using each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image and combining the name of the multimedia content to generate a promotional image of the multimedia content for display on the client includes: Obtaining a text of the name of the multimedia content in the language format required by the client; Each to-be-used material is used as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and combined with the acquired text, a promotional image of the multimedia content for display on the client is generated.

10. The method according to any one of claims 1 to 9, characterized in that: The number of the history tags is multiple, and the categories of the history tags include at least one of the following categories: a theme preference feature for representing the content classification of the account preference, a role preference feature for representing the content role of the account preference, and a plot preference feature for representing the content plot of the account preference.

11. A promotional image generating device, characterized in that: Applied to a server, the device comprises: A receiving module, used for receiving a request for obtaining a promotional image of multimedia content sent by a client; An acquisition module, used to acquire the history tags of the target account that logs into the client, and to acquire the material library corresponding to the multimedia content; wherein one material library corresponds to one type of element in the promotional image, and different material libraries correspond to different types of elements; A determination module, used to determine the material matching the historical tag from each material library as the to-be-used material corresponding to the material library; The generating module is used to use each to-be-used material as an element of a type corresponding to the material library to which the to-be-used material belongs in the promotional image, and to generate a promotional image of the multimedia content for display on the client.

12. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing any of the methods described in claims 1-10 when executing a program stored in a memory.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

14. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 10 when being executed by a processor.

Citation Information

Patent Citations

  • Image synthesis method and device, equipment and storage medium

    CN114241073A

  • Image generation method and device, storage medium and computer equipment

    CN115409917A

  • Image generation method and device, equipment, storage medium and program product

    CN116127102A

  • Mobile terminal Banner picture generation method, equipment and medium

    CN116843791A

  • Charging device for provide power based on USB-PD and method of operating the same

    KR1020250009329A