Content recommendation method and apparatus, and storage medium and electronic device

Abstract

The present application relates to the technical field of artificial intelligence. Disclosed are a content recommendation method and apparatus, and a storage medium and an electronic device. The method comprises: acquiring a multi-dimensional prompt language; inputting the multi-dimensional prompt language into a preset mask language model, so as to obtain a predicted term corresponding to a mask in a preset template statement; obtaining recommendation information on the basis of the predicted term; and performing content recommendation on the basis of the recommendation information. By means of the present application, the content recommendation effect can be effectively improved, and the content recommendation cost can also be reduced.

Description

Content recommendation method, device, storage medium and electronic device

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on September 19, 2023, with application number 202311215260.6 and application name “Content recommendation method, device, storage medium and electronic device”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of artificial intelligence technology, and specifically to a content recommendation method, device, storage medium, and electronic device. Background Art

[0003] Taking the content scenario of film and television recommendation as an example, content recommendation is usually made based on multi-dimensional information, such as poster dimension, user portrait dimension, etc.

[0004] Current content recommendation solutions typically convert information of different dimensions into corresponding features, such as converting posters into labels and portraits into numerical features. These dimensions are then fed into the model in different forms for recommendation analysis. This approach often results in asymmetric representation of information across dimensions, leading to poor recommendation analysis and content recommendation effectiveness. Furthermore, the model requires accumulating a large amount of training data from scratch and relying on significant computing power for training, resulting in high content recommendation costs. Technical issues

[0005] Under the current content recommendation method, there are problems such as poor content recommendation effect and high content recommendation cost, which affects user experience. Technical Solutions

[0006] The embodiments of the present application provide a solution that can effectively improve the content recommendation effect while reducing the content recommendation cost.

[0007] The embodiments of this application provide the following technical solutions:

[0008] According to one embodiment of the present application, a content recommendation method includes: obtaining multi-dimensional prompt language corresponding to a target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; inputting the multi-dimensional prompt language into a preset mask language model for language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; obtaining recommendation information based on the predicted words corresponding to the mask in the preset template sentence; and recommending content to the target user based on the recommendation information.

[0009] In some embodiments of the present application, the preset masked language model is trained in the following manner: a training data set is obtained, the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample masked sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimensional information of each dimension; with the sample words corresponding to the mask in the sample masked sentences in each group of the training data as the expected output, the pre-trained language model is trained to perform language mask prediction analysis based on the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset masked language model.

[0010] In some embodiments of the present application, the multiple dimensions include content dimension, user dimension and auxiliary dimension; the prompt language of the content dimension, user dimension and auxiliary dimension is generated in the following manner: obtaining the dimensional information of the content dimension, and converting the dimensional information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension; obtaining the dimensional information of the user dimension, and converting the dimensional information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension; obtaining the dimensional information of the auxiliary dimension, and converting the dimensional information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension.

[0011] In some embodiments of the present application, the content dimension includes a poster dimension; converting the dimensional information of the content dimension into a natural language sequence to obtain a prompt language for the content dimension includes: obtaining dimensional information of the poster dimension, the dimensional information of the poster dimension including the poster of the candidate content; extracting information from the poster to obtain a poster description statement of the poster; generating a natural language sequence of the poster dimension according to the poster description statement to obtain a prompt language for the poster dimension.

[0012] In some embodiments of the present application, a natural language sequence of the poster dimension is generated based on the poster description statement, including one of the following methods: using the poster description statement as the natural language sequence of the poster dimension; enhancing the poster description statement to obtain an enhanced description statement, and organizing the enhanced description statement to obtain a natural language sequence of the poster dimension.

[0013] In some embodiments of the present application, the content dimension includes a media asset dimension; converting the dimensional information of the content dimension into a natural language sequence to obtain a prompt language for the content dimension includes: obtaining dimensional information of the media asset dimension, the dimensional information of the media asset dimension including media asset information of candidate content; converting the media asset information of the candidate content into a natural language sequence to obtain a prompt language for the media asset dimension.

[0014] In some embodiments of the present application, the user dimension includes a selection dimension; converting the dimensional information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension includes: obtaining the content preference selection information selected by the target user himself; converting the content preference selection information into a natural language sequence to obtain the prompt language of the selection dimension.

[0015] In some embodiments of the present application, the user dimension may include a portrait dimension; converting the dimensional information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension includes: obtaining the portrait information of the target user; converting the portrait information into a natural language sequence to obtain the prompt language of the portrait dimension.

[0016] In some embodiments of the present application, the auxiliary dimension includes a scene dimension; converting the dimensional information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension includes: obtaining one or more of the device information, time information, emotional information, and weather information corresponding to the target user in real time to obtain the dimensional information of the scene dimension; converting the dimensional information of the scene dimension into a natural language sequence to obtain the prompt language of the scene dimension.

[0017] According to one embodiment of the present application, a content recommendation device includes: an acquisition module for acquiring multi-dimensional prompt language corresponding to a target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; an analysis module for inputting the multi-dimensional prompt language into a preset mask language model for language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; a processing module for obtaining recommendation information based on the predicted words corresponding to the mask in the preset template sentence; and a recommendation module for recommending content to the target user based on the recommendation information.

[0018] In some embodiments of the present application, the device includes a training module, which is used to: obtain a training data set, wherein the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample mask sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimensional information of each dimension; with the sample words corresponding to the mask in the sample mask sentences in each group of the training data as the expected output, train a pre-trained language model to perform language mask prediction analysis based on the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset masked language model.

[0019] In some embodiments of the present application, the multiple dimensions include content dimension, user dimension and auxiliary dimension; the device includes a generation module, which is used to: obtain dimension information of the content dimension, and convert the dimension information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension; obtain dimension information of the user dimension, and convert the dimension information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension; obtain dimension information of the auxiliary dimension, and convert the dimension information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension.

[0020] In some embodiments of the present application, the content dimension includes a poster dimension; the generation module is used to: obtain dimensional information of the poster dimension, the dimensional information of the poster dimension includes the poster of the candidate content; extract information from the poster to obtain a poster description statement of the poster; generate a natural language sequence of the poster dimension according to the poster description statement to obtain a prompt language of the poster dimension.

[0021] In some embodiments of the present application, the generation module is used to implement one of the following methods: using the poster description statement as a natural language sequence of the poster dimension; enhancing the poster description statement to obtain an enhanced description statement, and organizing the enhanced description statement to obtain a natural language sequence of the poster dimension.

[0022] In some embodiments of the present application, the content dimension includes a media asset dimension; the generation module is used to: obtain dimensional information of the media asset dimension, the dimensional information of the media asset dimension includes media asset information of the candidate content; convert the media asset information of the candidate content into a natural language sequence to obtain a prompt language of the media asset dimension.

[0023] In some embodiments of the present application, the user dimension includes a selection dimension; the generation module is used to: obtain content preference selection information selected by the target user himself; convert the content preference selection information into a natural language sequence to obtain a prompt language for the selection dimension.

[0024] In some embodiments of the present application, the user dimension may include a portrait dimension; the generation module is used to: obtain the portrait information of the target user; convert the portrait information into a natural language sequence to obtain a prompt language for the portrait dimension.

[0025] In some embodiments of the present application, the auxiliary dimension includes a scene dimension; the generation module is used to: obtain one or more of the device information, time information, emotional information, and weather information corresponding to the target user in real time to obtain dimensional information of the scene dimension; convert the dimensional information of the scene dimension into a natural language sequence to obtain a prompt language of the scene dimension.

[0026] According to another embodiment of the present application, a storage medium stores a computer program thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method described in the embodiment of the present application.

[0027] According to another embodiment of the present application, an electronic device may include: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the method described in the embodiment of the present application.

[0028] According to another embodiment of the present application, a computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations described in the embodiments of the present application. Beneficial effects

[0029] In an embodiment of the present application, a multi-dimensional prompt language corresponding to a target user is obtained, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; the multi-dimensional prompt language is input into a preset mask language model for language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; based on the predicted words corresponding to the mask in the preset template sentence, recommendation information is obtained; and content is recommended to the target user based on the recommendation information.

[0030] In this way, the natural language sequence obtained by converting the multi-dimensional dimensional information is used as the multi-dimensional prompt language, so that the multi-dimensional dimensional information can be effectively symmetrical. The multi-dimensional prompt language is further input into the preset mask language model for language mask prediction analysis to obtain the predicted words corresponding to the mask in the preset template sentence. The predicted words can effectively reflect the user's preference for the content. The recommendation task can be effectively completed based on the predicted words, and the implementation method of the recommendation task is converted into a natural language mask understanding task. The preset mask language model can be trained with small samples, high performance and low cost, which effectively improves the content recommendation effect as a whole, while reducing the content recommendation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0032] FIG1 shows a flowchart of a content recommendation method according to an embodiment of the present application.

[0033] FIG2 shows an interface diagram of extracting sentences from a poster according to an example of the present application.

[0034] FIG3 shows an interface diagram of a user selecting preferences according to an example of the present application.

[0035] FIG4 shows a schematic diagram of model training and application in a scenario according to the present application.

[0036] FIG5 shows a block diagram of a content recommendation device according to an embodiment of the present application.

[0037] FIG6 shows a block diagram of an electronic device according to an embodiment of the present application.

[0038] Implementation Methods of the Application

[0039] The present disclosure will be further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the examples provided herein are merely for explaining the present disclosure and are not intended to limit the present disclosure. In addition, the examples provided below are partial examples for implementing the present disclosure, rather than providing all examples for implementing the present disclosure. In the absence of conflict, the technical solutions described in the examples of the present disclosure may be implemented in any combination.

[0040] It should be noted that, in the embodiments of the present disclosure, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a method or apparatus comprising a series of elements includes not only the elements explicitly stated, but also other elements not explicitly listed, or also includes elements inherent to the implementation of the method or apparatus. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other related elements (such as steps in the method or units in the apparatus, for example, a unit may be part of a circuit, part of a processor, part of a program or software, etc.) in the method or apparatus comprising the element.

[0041] For example, the content recommendation method provided by the embodiment of the present disclosure includes a series of steps, but the content recommendation method provided by the embodiment of the present disclosure is not limited to the recorded steps. Similarly, the content recommendation device provided by the embodiment of the present disclosure includes a series of units, but the device provided by the embodiment of the present disclosure is not limited to including the units explicitly recorded, and may also include units required to obtain relevant information or perform processing based on the information.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure pertains. The terms used herein are for the purpose of describing specific embodiments only and are not intended to limit the present disclosure.

[0043] Figure 1 schematically illustrates a flow chart of a content recommendation method according to an embodiment of the present application. The content recommendation method may be executed by a terminal or a server, where the terminal may be, for example, a television, computer, mobile phone, smartwatch, or home appliance, and the server may be, for example, a cloud server or a physical server.

[0044] As shown in FIG1 , the content recommendation method may include steps S110 to S140 .

[0045] Step S110, obtaining the multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; Step S120, inputting the multi-dimensional prompt language into a preset mask language model for language mask prediction analysis to obtain the predicted words corresponding to the mask in the preset template sentence; Step S130, obtaining recommendation information based on the predicted words corresponding to the mask in the preset template sentence; Step S140, recommending content to the target user based on the recommendation information.

[0046] The target user is the user to whom content is being recommended. A target user can correspond to an account, for example, an account logged into an app, which can belong to the target user. Multiple dimensions can include at least two of the following: scene dimension, selection dimension, portrait dimension, content dimension, and poster dimension. Each dimension can have a corresponding prompt language, which is specifically a natural language sequence converted from the dimension information. Multiple dimensions can be set based on actual circumstances.

[0047] The prompt language of each dimension can be pre-converted offline and stored in a predetermined location, and the prompt language of each dimension can be obtained from the predetermined location; the prompt language of each dimension can be obtained by real-time online conversion. By obtaining the dimensional information of the target user in each dimension in real time, and converting the dimensional information into a natural language sequence in real-time online, the prompt language of each dimension can be obtained.

[0048] The preset masked language model is a pre-trained deep learning model used for language mask prediction analysis. By inputting multi-dimensional prompt language into the preset masked language model for language mask prediction analysis, the model outputs predicted words corresponding to the masked words in the preset template sentence. The predicted words can reflect the target user's preferences for a particular piece of content.

[0049] For example, the preset template sentence is "This video is [mask] to me", where [mask] is the mask in the preset template sentence. The multi-dimensional prompt language is input into the preset mask language model for language mask prediction analysis, and the predicted words corresponding to [mask] can be obtained, such as Amazing.

[0050] Based on the predicted words corresponding to the mask in the preset template sentence, recommendation information can be obtained. The recommendation information can describe whether to recommend a certain content to the target user and / or how to recommend a certain content to the target user. For example, if the predicted word is a word of type A, the recommendation information can be not recommended; if the predicted word is a word of type B, the recommendation information can be recommended; if the predicted word is a word of type C, the recommendation information can be the content recommendation method corresponding to type C (such as recommendation duration, recommendation strength, recommendation location, etc.); if the predicted word is a word of type D, the recommendation information can be the content recommendation method corresponding to type D.

[0051] Furthermore, content can be recommended to the target user based on the recommendation information, for example, a certain content can be recommended to the target user, or a certain content can be recommended to the target user according to a determined content recommendation method.

[0052] In this way, based on steps S110 to S140, the natural language sequence obtained by converting the multi-dimensional dimensional information is used as the multi-dimensional prompt language, so that the multi-dimensional dimensional information can be effectively symmetrical, and the multi-dimensional prompt language is further input into the preset mask language model for language mask prediction analysis to obtain the predicted words corresponding to the mask in the preset template sentence. The predicted words can effectively reflect the user's preference for the content. The recommendation task can be effectively completed based on the predicted words, and the implementation method of the recommendation task is converted into a natural language mask understanding task. The preset mask language model can be trained with small samples, high performance and low cost, which effectively improves the content recommendation effect as a whole, while reducing the content recommendation cost.

[0053] The following describes further optional specific embodiments of each step performed when performing content recommendation under the embodiment of Figure 1.

[0054] Step S110 : obtaining a multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension.

[0055] The prompt language of each dimension can be pre-converted offline and stored in a predetermined location, and the prompt language of each dimension can be obtained from the predetermined location; the prompt language of each dimension can be obtained by real-time online conversion. By obtaining the dimensional information of the target user in each dimension in real time, and converting the dimensional information into a natural language sequence in real-time online, the prompt language of each dimension can be obtained.

[0056] In one embodiment, the multiple dimensions may include a content dimension, a user dimension, and an auxiliary dimension; and the prompt language of the content dimension, the user dimension, and the auxiliary dimension may be generated in the following manner:

[0057] Acquire the dimension information of the content dimension, and convert the dimension information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension; acquire the dimension information of the user dimension, and convert the dimension information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension; acquire the dimension information of the auxiliary dimension, and convert the dimension information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension.

[0058] The content dimension refers to the dimension corresponding to the candidate content to be recommended, the user dimension refers to the dimension corresponding to the target user, and the auxiliary dimension refers to the dimension other than the content dimension and the user dimension. The content dimension can include at least one of the poster dimension and the media asset dimension. The user dimension can include at least one of the portrait dimension and the selection dimension.

[0059] The dimensional information of the content dimension, user dimension and auxiliary dimension are respectively converted into natural language sequences to obtain the prompt languages ​​of the content dimension, user dimension and auxiliary dimension.

[0060] By integrating the prompt languages ​​of the three dimensions of content dimension, user dimension and auxiliary dimension, the preset mask language model can perform language mask prediction analysis considering the prompt languages ​​of these three dimensions, and accurately obtain predicted words that effectively reflect the user's preferences for the candidate content.

[0061] Furthermore, in one embodiment, the content dimension includes a poster dimension; converting the dimensional information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension includes: obtaining dimensional information of the poster dimension, the dimensional information of the poster dimension includes the poster of the candidate content; extracting information from the poster to obtain a poster description statement of the poster; generating a natural language sequence of the poster dimension according to the poster description statement to obtain the prompt language of the poster dimension.

[0062] Referring to Figure 2, for candidate content (e.g., movies, TV series, items, etc.), corresponding posters can be obtained. Information extraction is performed on the poster shown on the left side of Figure 2 to obtain a poster description sentence for the poster shown on the right side of Figure 2. Furthermore, a natural language sequence in the poster dimension can be generated based on the poster description sentence. The natural language sequence in the poster dimension is the resulting prompt language in the poster dimension.

[0063] Posters are often the first cultural symbol users encounter, offering the most appeal and appeal. They transcend age, language, and nationality, allowing people to perceive the intended message of the work. Using the poster's dimensional language can effectively analyze candidate content recommendations.

[0064] The information extraction from the poster to obtain the poster description statement can be specifically performed by inputting the poster into a preset poster expression model, and the preset poster expression model outputting the poster description statement based on the poster. The preset poster expression model is preferably a pre-trained deep learning model specifically designed to convert poster extraction into poster description statements. The preset poster expression model can also be a selected existing large model.

[0065] Furthermore, generating the natural language sequence of the poster dimension according to the poster description sentence may specifically include one of the following methods:

[0066] The first method is to use the poster description sentence as the natural language sequence of the poster dimension; the second method is to enhance the poster description sentence to obtain an enhanced description sentence, and then organize the enhanced description sentence to obtain a natural language sequence of the poster dimension.

[0067] In the second method, the poster description is enhanced to obtain an enhanced description. Specifically, the poster is compared with a preset reference poster to obtain poster comparison information; the poster description and the poster comparison information are combined to obtain the enhanced description. The enhanced description is then organized to obtain a poster-specific natural language sequence. Specifically, the enhanced description is organized through lemmatization, stemming, duplicate word removal, keyword extraction, and keyword-based sentence formation to obtain a poster-specific natural language sequence.

[0068] The preset reference posters can be those whose browsing patterns among target users meet predefined collection criteria (e.g., highest browsing frequency). Poster comparison information can include color, character, and text differences. Through comparison enhancement, the resulting poster-dimensional natural language sequence (also known as prompt information) can further improve the accuracy of predicted words and enhance content recommendations.

[0069] Furthermore, in one embodiment, the content dimension includes a media asset dimension; converting the dimensional information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension includes: obtaining dimensional information of the media asset dimension, the dimensional information of the media asset dimension includes media asset information of candidate content; converting the media asset information of the candidate content into a natural language sequence to obtain the prompt language of the media asset dimension.

[0070] For example, as shown in the following table, the media asset dimension may include sub-media asset dimensions such as click volume and completion rate, and the media asset information of the candidate content may include the media asset information of each sub-media asset dimension under the media asset dimension.

[0071] The media asset information of the candidate content is converted into a natural language sequence to obtain the prompt language of the media asset dimension. Specifically, the media asset information of each sub-media asset dimension can be converted into a prompt sentence, and then the prompt sentence of each sub-media asset dimension is generated into a natural language sequence to obtain the prompt language of the media asset dimension.

[0072] For example, "This video is a 4K high-resolution Wuxia film, produced by C1. It has an 8.3 rating and is free for M1 members. It's trending in the last 48 hours and has a 68% audience retention rate. It features actress LIU and four celebrities in this movie."

[0073] The method for converting the media asset information of each sub-asset dimension into a prompt statement can specifically include inputting the media asset information of each sub-asset dimension into a preset media asset expression model, which then outputs a prompt statement based on the media asset information of each sub-asset dimension. The preset media expression model is preferably a pre-trained deep learning model specifically designed to convert the media information of each sub-asset dimension into a prompt statement. The preset media expression model can also be a large existing model.

[0074] Alternatively, in other embodiments, the media asset information of each sub-asset dimension may be converted into a prompt statement by entering the media asset information of each sub-asset dimension into a preset media asset language sequence template to obtain a prompt statement. For example, an example media language sequence template may be "This video is a*film," and entering the media asset information of the sub-asset dimension into the preset media language sequence template may result in a prompt statement such as "This video is a wuxia film."

[0075] In one embodiment, the user dimension may include a selection dimension; converting the dimension information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension may specifically be: obtaining the content preference selection information selected by the target user himself; converting the content preference selection information into a natural language sequence to obtain the prompt language of the selection dimension.

[0076] Content preference selection information can be the target user's selection preference information for the content selected. Content preference selection information refers to dimensional information about the selection dimension. As shown in FIG3 , the target user can select one or more of "Latest," "Popular," and so on in a preset content preference selection page as shown in FIG3 to obtain selection preference information. It is understood that selection preference information can also include other user-selectable information set based on actual circumstances.

[0077] Because this solution utilizes a pre-set masked language model, it captures the target user's own content preferences, increasing their autonomy and allowing them to determine which content preferences should be considered by the recommendation system. This effectively counteracts the social biases or stereotypes that pre-set masked language models may impose on different demographics. For example, for a three-year-old girl, the model might assume she likes Barbie videos, leading to a high recommendation load of similar videos and reduced exposure to other types of videos. This leads to a high concentration of homogenous media resources, which in turn exacerbates the biases introduced by the model. If a user chooses to watch a recommended video because they can't find other videos, the resulting user profile will also be biased by social attributes, further deteriorating the recommendation results.

[0078] In particular, the preset masked language model is obtained by training with a pre-trained language model. The pre-trained language model inevitably carries more or less social biases or stereotypes of different groups of people. By obtaining the content preference selection information of the target users themselves, the target users' autonomous selection space is increased, the bias can be effectively combated, and the content recommendation effect is improved.

[0079] The target user can individually select content preferences in real time on a preset content preference selection page, thereby obtaining content preference selection information. This content preference selection information can then be converted into a natural language sequence in real time. The converted natural language sequence serves as the prompt language for the selection dimension. This means that the prompt language for the selection dimension can be generated in real time. Alternatively, the target user can also individually select content preferences in advance on a preset content preference selection page, thereby obtaining content preference selection information. This content preference selection information can then be converted into a natural language sequence in advance. The converted natural language sequence serves as the prompt language for the selection dimension. This means that the prompt language for the selection dimension can be generated offline.

[0080] In one approach, content preference selection information is converted into a natural language sequence. Specifically, the content preference selection information for a selection dimension is input into a preset selection expression model, which then outputs a natural language sequence for the selection dimension based on the content preference selection information for the selection dimension. The preset selection expression model is preferably a pre-trained deep learning model specifically designed to convert dimensional information for the selection dimension into a natural language sequence. The preset selection expression model may also be a pre-existing large model.

[0081] Alternatively, in another embodiment, the content preference selection information is converted into a natural language sequence. Specifically, the content preference selection information may be entered into a preset selection language sequence template to obtain a natural language sequence for the selection dimension. For example, an example selection language sequence template may be "My personal preference for *video," and the natural language sequence obtained by entering the content preference selection information into the template may be "My personal preference for free video."

[0082] In one embodiment, the user dimension may include a portrait dimension; converting the dimensional information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension may specifically be: obtaining the portrait information of the target user; converting the portrait information into a natural language sequence to obtain the prompt language of the portrait dimension.

[0083] For example, as shown in the following table, the portrait dimension may include sub-portrait dimensions such as member information and age, and the portrait information may include media resource information of each sub-portrait dimension under the portrait dimension.

[0084] The portrait information is converted into a natural language sequence to obtain the prompt language of the portrait dimension. Specifically, the portrait information of each sub-portrait dimension can be converted into a prompt sentence, and then the prompt sentence of each sub-portrait dimension is generated into a natural language sequence to obtain the prompt language of the portrait dimension.

[0085] The prompt language of the portrait dimension is for example I am a teenage girl.I have M2 and M3 membership.I like G style cartoon and animation and dislikeG1 videos.

[0086] The portrait information of each sub-portrait dimension can be converted into a prompt statement by, for example, inputting the portrait information of each sub-portrait dimension into a preset portrait expression model, which then outputs a prompt statement based on the portrait information of each sub-portrait dimension. The preset portrait expression model is preferably a pre-trained deep learning model specifically designed to convert the portrait information of each sub-portrait dimension into a prompt statement. The preset portrait expression model can also be an existing large model.

[0087] Alternatively, in other embodiments, the portrait information of each sub-portrait dimension may be converted into a prompt statement by entering the portrait information of each sub-portrait dimension into a preset portrait language sequence template to obtain a prompt statement. For example, an example portrait language sequence template may be "I am is a*", and the prompt statement obtained by entering the portrait information of each sub-portrait dimension into the preset portrait language sequence template may be "I am a girl".

[0088] In one embodiment, the auxiliary dimension may include a scene dimension; converting the dimensional information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension may specifically be: obtaining one or more of the device information, time information, emotional information, and weather information corresponding to the target user in real time to obtain the dimensional information of the scene dimension; converting the dimensional information of the scene dimension into a natural language sequence to obtain the prompt language of the scene dimension.

[0089] Device information refers to information about the device used by the target user, including device type and model. Time information refers to real-time time information, such as year, month, day, and hour. Device information and time information can be obtained in real time from the terminal currently used by the target user.

[0090] Emotional information refers to the target user's emotions, such as happiness or sadness. This information can be obtained through real-time facial recognition or voiceprint recognition of the target user, or it can be manually input or selected by the user in real time. For example, the target user's emotions can be identified by their speech rate, volume, pitch, amplitude, pitch frequency, and duration, or by identifying their emotions in commands they enter.

[0091] Weather information refers to weather information corresponding to the target user, such as whether it will rain, temperature, humidity, pollution level, etc. Weather information can be obtained in real time from the terminal used by the target user, or can be obtained from a weather forecast platform based on the location of the terminal used by the target user.

[0092] It is understandable that dimensional information of scene dimensions such as device information, time information, emotion information, weather information, etc. can be obtained through any agreed legal method, and this application does not make specific restrictions on this.

[0093] The dimensional information of the scene dimension is usually collected in real time when content recommendation is needed for the target user. After obtaining the dimensional information of the scene dimension in real time (one or more of the target user's real-time corresponding device information, time information, emotional information, and weather information), the dimensional information of the scene dimension is converted into a natural language sequence. The natural language sequence of the scene dimension obtained by conversion is the prompt language of the scene dimension.

[0094] In one approach, the dimensional information of the scene dimension is converted into a natural language sequence. Specifically, the dimensional information of the scene dimension can be input into a preset scene expression model, and the preset scene expression model outputs a natural language sequence of the scene dimension based on the dimensional information of the scene dimension. The preset scene expression model is preferably a pre-trained deep learning model specifically designed to convert the dimensional information of the scene dimension into a natural language sequence. The preset scene expression model can also be a large existing model.

[0095] Alternatively, in other methods, the dimensional information of the scenario dimension is converted into a natural language sequence. Specifically, the dimensional information of the scenario dimension may be entered into a preset scenario language sequence template to obtain the natural language sequence of the scenario dimension. For example, an example scenario language sequence template may be "It is 'N' o'clock in 'W'. I feel 'E'". After entering the dimensional information of the scenario dimension, the resulting natural language sequence may be "It is 18:00 o'clock in workdays, I feel exhausted".

[0096] Step S120 : Input the multi-dimensional prompt language into a preset mask language model to perform language mask prediction analysis to obtain predicted words corresponding to the mask in the preset template sentence.

[0097] The preset masked language model is a pre-trained deep learning model used for language mask prediction analysis. By inputting multi-dimensional prompt language into the preset masked language model for language mask prediction analysis, the model outputs predicted words corresponding to the masked words in the preset template sentence. The predicted words can reflect the target user's preferences for a particular piece of content.

[0098] For example, the preset template sentence is "This video is [mask] to me", where [mask] is the mask in the preset template sentence. The multi-dimensional prompt language is input into the preset mask language model for language mask prediction analysis, and the predicted words corresponding to [mask] can be obtained, such as Amazing.

[0099] In one embodiment, the preset masked language model is trained in the following manner: a training data set is obtained, the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample masked sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimensional information of each dimension; taking the masked sample words corresponding to the mask in the sample masked sentences in each group of the training data as the expected output, the pre-trained language model is trained to perform language mask prediction analysis based on the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset masked language model.

[0100] Each set of training data includes multi-dimensional sample language, corresponding sample masked sentences, and sample words. The multi-dimensional sample language can include sample language from the content dimension, user dimension, and auxiliary dimension. The sample language serves as the sample prompt language. A sample masked sentence, for example, is "This video is [mask] to me." The sample word is the word label corresponding to the mask [mask] in the sample masked sentence, such as "Awful."

[0101] In one example, content can be categorized into eight types of sample words based on its click-through rate (CTR), conversion rate (CPC), and completion rate (ARR). The predicted word output by the preset masked language model is one of these eight types of sample words. These eight types of sample words include, for example, "Awful," "Potential," "Classical," "Niche," "Good," "Amazing," "Clickbait," and "Fantastic." In other examples, sample words could also include "will," "won't," or "dislike," which can express user preferences.

[0102] The pre-trained language model can be a Bert model, a SpanBert model, or other pre-trained language models. Bert models include models such as bert_base_uncased and RoBerta-large. Taking the sample words corresponding to the mask in the sample masked sentences in each set of training data as the desired output, the pre-trained language model is trained to perform language mask prediction analysis based on the multi-dimensional sample language in each set of training data. For example, the multi-dimensional sample language in the training data is input into the pre-trained language model so that the pre-trained language model predicts the sample words corresponding to the mask in the sample masked sentences. The parameters in the pre-trained language model are continuously adjusted based on the prediction accuracy until the predetermined training conditions are met, thereby obtaining a trained pre-trained language model, i.e., a preset masked language model. The predetermined training conditions may include a prediction accuracy higher than a predetermined accuracy rate or a predetermined number of training times, etc.

[0103] In this embodiment, the pre-trained language model is further trained to obtain a preset masked language model, which can be used for content recommendation in an efficient and low-cost manner. It is understood that in other embodiments, the preset masked language model can be obtained based on other neural network training.

[0104] In one scenario, referring to FIG4 , as shown in the “Offline” portion on the left side of FIG4 , the multi-dimensional sample language in each set of training data specifically includes sample language (prompt) corresponding to the scene dimension, poster dimension, media asset dimension, portrait dimension, and selection dimension. The pre-trained language model is specifically a Bert model. With the sample words corresponding to the mask in the sample masked sentences in each set of training data as the expected output, the Bert model is trained to perform language mask prediction analysis based on the multi-dimensional sample language in each set of training data until the predetermined training conditions are met, thereby obtaining a preset masked language model (Bert-mlm model).

[0105] Furthermore, as shown in the "Online" section on the right side of Figure 4, when recommending target content, the target user's multi-dimensional prompt language is obtained. The multi-dimensional prompt language includes prompts corresponding to the scene dimension, poster dimension, media resource dimension, portrait dimension, and selection dimension. The prompts corresponding to the scene dimension, poster dimension, media resource dimension, portrait dimension, and selection dimension are input into the preset masked language model (Bert-mlm model) for language mask prediction analysis to obtain the predicted words corresponding to the mask in the preset template sentence.

[0106] To facilitate better implementation of the content recommendation method provided in the embodiments of this application, the embodiments of this application also provide a content recommendation device based on the aforementioned content recommendation method. The meanings of the terms herein are the same as those in the aforementioned content recommendation method. For specific implementation details, please refer to the description in the method embodiments. Figure 5 shows a block diagram of a content recommendation device according to one embodiment of the present application.

[0107] As shown in Figure 5, the content recommendation device 200 may include: an acquisition module 210 can be used to obtain multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; an analysis module 220 can be used to input the multi-dimensional prompt language into a preset masked language model for language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; a processing module 230 can be used to obtain recommendation information based on the predicted words corresponding to the mask in the preset template sentence; and a recommendation module 240 can be used to recommend content to the target user based on the recommendation information.

[0108] In some embodiments of the present application, the device includes a training module, which is used to: obtain a training data set, wherein the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample mask sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimensional information of each dimension; with the sample words corresponding to the mask in the sample mask sentences in each group of the training data as the expected output, train a pre-trained language model to perform language mask prediction analysis based on the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset masked language model.

[0109] In some embodiments of the present application, the multiple dimensions include content dimension, user dimension and auxiliary dimension; the device includes a generation module, which is used to: obtain dimension information of the content dimension, and convert the dimension information of the content dimension into a natural language sequence to obtain the prompt language of the content dimension; obtain dimension information of the user dimension, and convert the dimension information of the user dimension into a natural language sequence to obtain the prompt language of the user dimension; obtain dimension information of the auxiliary dimension, and convert the dimension information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension.

[0110] In some embodiments of the present application, the content dimension includes a poster dimension; the generation module is used to: obtain dimensional information of the poster dimension, the dimensional information of the poster dimension includes the poster of the candidate content; extract information from the poster to obtain a poster description statement of the poster; generate a natural language sequence of the poster dimension according to the poster description statement to obtain a prompt language of the poster dimension.

[0111] In some embodiments of the present application, the generation module is used to implement one of the following methods: using the poster description statement as a natural language sequence of the poster dimension; enhancing the poster description statement to obtain an enhanced description statement, and organizing the enhanced description statement to obtain a natural language sequence of the poster dimension.

[0112] In some embodiments of the present application, the content dimension includes a media asset dimension; the generation module is used to: obtain dimensional information of the media asset dimension, the dimensional information of the media asset dimension includes media asset information of the candidate content; convert the media asset information of the candidate content into a natural language sequence to obtain a prompt language of the media asset dimension.

[0113] In some embodiments of the present application, the user dimension includes a selection dimension; the generation module is used to: obtain content preference selection information selected by the target user himself; convert the content preference selection information into a natural language sequence to obtain a prompt language for the selection dimension.

[0114] In some embodiments of the present application, the user dimension may include a portrait dimension; the generation module is used to: obtain the portrait information of the target user; convert the portrait information into a natural language sequence to obtain a prompt language for the portrait dimension.

[0115] In some embodiments of the present application, the auxiliary dimension includes a scene dimension; the generation module is used to: obtain one or more of the device information, time information, emotional information, and weather information corresponding to the target user in real time to obtain dimensional information of the scene dimension; convert the dimensional information of the scene dimension into a natural language sequence to obtain a prompt language of the scene dimension.

[0116] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0117] In addition, an embodiment of the present application further provides an electronic device, as shown in FIG6 , which shows a schematic structural diagram of the electronic device involved in the embodiment of the present application. Specifically:

[0118] The electronic device may include one or more processing cores of the processor 301, one or more computer-readable storage media of the memory 302, a power supply 303 and other components. Those skilled in the art will understand that the electronic device structure shown in FIG6 does not constitute a limitation of the electronic device, and may include more or fewer components than shown, or combine certain components, or arrange the components differently. Among them:

[0119] Processor 301 is the control center of the electronic device, connecting the various components of the entire computer device using various interfaces and lines. By running or executing software programs and / or modules stored in memory 302 and accessing data stored in memory 302, it performs various computer device functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, processor 301 may include one or more processing cores; preferably, processor 301 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 301.

[0120] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and data processing by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 302 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0121] The electronic device also includes a power supply 303 for supplying power to various components. Preferably, the power supply 303 can be logically connected to the processor 301 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 303 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0122] Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to one or more computer program processes into the memory 302 according to the following instructions, and the processor 301 runs the computer program stored in the memory 302, thereby implementing the various functions of the aforementioned embodiments of the present application. For example, the processor 301 may perform the following steps:

[0123] Obtain multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; input the multi-dimensional prompt language into a preset mask language model for language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; obtain recommendation information based on the predicted words corresponding to the mask in the preset template sentence; and recommend content to the target user based on the recommendation information.

[0124] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by a computer program, or by controlling related hardware through a computer program. The computer program may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0125] To this end, an embodiment of the present application further provides a storage medium storing a computer program, which can be loaded by a processor to execute the steps of any method provided in the embodiment of the present application.

[0126] The storage medium may be a computer-readable storage medium, which may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0127] Since the computer program stored in the storage medium can execute the steps of any method provided in the embodiments of the present application, the beneficial effects that can be achieved by the method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0128] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0129] It should be understood that the present application is not limited to the embodiments that have been described above and shown in the accompanying drawings, but various modifications and changes may be made without departing from the scope thereof.

Claims

1. A content recommendation method, wherein: include: Acquire a multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; Inputting the multi-dimensional prompt language into a preset mask language model to perform language mask prediction analysis to obtain predicted words corresponding to the mask in the preset template sentence; Obtaining recommendation information according to the predicted words corresponding to the mask in the preset template sentence; Content is recommended to the target user according to the recommendation information.

2. The method according to claim 1, wherein: The preset mask language model is trained in the following manner: Acquire a training data set, wherein the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample mask sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimension information of each dimension; Taking the sample words corresponding to the masks in the sample masked sentences in each group of the training data as the expected output, the pre-trained language model is trained to perform language mask prediction analysis according to the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset masked language model.

3. The method according to any one of claims 1 to 2, wherein: The multiple dimensions include content dimension, user dimension and auxiliary dimension; the prompt language of the content dimension, user dimension and auxiliary dimension is generated in the following manner: Acquire dimension information of the content dimension, and convert the dimension information of the content dimension into a natural language sequence to obtain a prompt language of the content dimension; Acquire dimension information of the user dimension, and convert the dimension information of the user dimension into a natural language sequence to obtain a prompt language of the user dimension; The dimension information of the auxiliary dimension is obtained, and the dimension information of the auxiliary dimension is converted into a natural language sequence to obtain a prompt language of the auxiliary dimension.

4. The method according to claim 3, wherein: The content dimension includes a poster dimension; converting the dimension information of the content dimension into a natural language sequence to obtain a prompt language of the content dimension includes: Acquire dimension information of a poster dimension, where the dimension information of the poster dimension includes a poster of a candidate content; Extracting information from the poster to obtain a poster description statement of the poster; A natural language sequence of the poster dimension is generated according to the poster description sentence to obtain a prompt language of the poster dimension.

5. The method according to claim 4, wherein: The generating the natural language sequence of the poster dimension according to the poster description sentence includes one of the following methods: Taking the poster description sentence as a natural language sequence of the poster dimension; The poster description sentence is enhanced to obtain an enhanced description sentence, and the enhanced description sentence is sorted to obtain a natural language sequence of the poster dimension.

6. The method according to claim 3, wherein: The content dimension includes a media asset dimension; converting the dimension information of the content dimension into a natural language sequence to obtain a prompt language of the content dimension includes: Acquire dimension information of a media asset dimension, where the dimension information of the media asset dimension includes media asset information of the candidate content; The media asset information of the candidate content is converted into a natural language sequence to obtain a prompt language of the media asset dimension.

7. The method according to claim 3, wherein: The user dimension includes a selection dimension; and converting the dimension information of the user dimension into a natural language sequence to obtain a prompt language of the user dimension includes: Obtaining content preference selection information selected by the target user; The content preference selection information is converted into a natural language sequence to obtain the prompt language of the selection dimension.

8. The method according to claim 3, wherein: The user dimension may include a portrait dimension; and converting the dimension information of the user dimension into a natural language sequence to obtain a prompt language of the user dimension includes: Obtain the portrait information of the target user; convert the portrait information into a natural language sequence to obtain the prompt language of the portrait dimension.

9. The method according to claim 3, wherein: The auxiliary dimensions include scene dimensions; The converting the dimension information of the auxiliary dimension into a natural language sequence to obtain the prompt language of the auxiliary dimension includes: Obtain one or more of device information, time information, emotion information, and weather information corresponding to the target user in real time to obtain dimension information of the scene dimension; The dimension information of the scene dimension is converted into a natural language sequence to obtain a prompt language of the scene dimension.

10. The method according to claim 4, wherein: The extracting information from the poster to obtain a poster description statement of the poster includes: The poster is input into a preset poster expression model to obtain a poster description sentence output by the preset poster expression model.

11. The method according to claim 5, wherein: The step of enhancing the poster description sentence to obtain an enhanced description sentence includes: Comparing the poster with a preset reference poster to obtain poster comparison information; The poster description statement and the poster comparison information are combined to obtain an enhanced description statement.

12. The method according to claim 11, wherein: The poster comparison information includes, but is not limited to, one or more of color difference information, character difference information, and text difference information.

13. The method according to claim 6, wherein: The media asset information of the candidate content may include media asset information of each sub-media asset dimension under the media asset dimension; converting the media asset information of the candidate content into a natural language sequence to obtain the prompt language of the media asset dimension includes: Convert the media asset information of each sub-media asset dimension into a prompt statement; The prompt sentences of each sub-media asset dimension are converted into a natural language sequence to obtain the prompt language of the media asset dimension.

14. The method according to claim 13, wherein: The step of converting the media asset information of each sub-media asset dimension into a prompt statement includes: The media asset information of each sub-media asset dimension is input into a preset media asset expression model to obtain a prompt sentence output by the preset media asset expression model based on the media asset information of each sub-media asset dimension.

15. The method according to claim 13, wherein: The converting of the media asset information of each sub-media asset dimension into a prompt statement includes: The media asset information of each sub-media asset dimension is filled into the preset media asset language sequence template to obtain a prompt statement.

16. A content recommendation device, wherein: include: An acquisition module, used to acquire multi-dimensional prompt language corresponding to the target user, wherein the prompt language of each dimension is a natural language sequence obtained by converting the dimensional information of each dimension; An analysis module, used for inputting the multi-dimensional prompt language into a preset mask language model to perform language mask prediction analysis to obtain predicted words corresponding to the mask in a preset template sentence; A processing module, configured to obtain recommendation information according to the predicted words corresponding to the mask in the preset template sentence; A recommendation module is used to recommend content to the target user based on the recommendation information.

17. The device according to claim 16, wherein: The device includes a training module, which is used to: obtain a training data set, wherein the training data set includes multiple groups of training data, each group of the training data includes multi-dimensional sample language and corresponding sample mask sentences and sample words, wherein the sample language of each dimension is a natural language sequence obtained by converting the sample dimension information of each dimension; taking the sample words corresponding to the mask in the sample mask sentences in each group of the training data as the expected output, training a pre-trained language model to perform language mask prediction and analysis according to the multi-dimensional sample language in each group of the training data until the predetermined training conditions are met to obtain the preset mask language model.

18. The device according to claim 16, wherein: The multiple dimensions include content dimension, user dimension and auxiliary dimension; the device includes a generation module, which is used to: obtain dimension information of the content dimension, and convert the dimension information of the content dimension into a natural language sequence to obtain a prompt language of the content dimension; obtain dimension information of the user dimension, and convert the dimension information of the user dimension into a natural language sequence to obtain a prompt language of the user dimension; obtain dimension information of the auxiliary dimension, and convert the dimension information of the auxiliary dimension into a natural language sequence to obtain a prompt language of the auxiliary dimension.

19. A storage medium, wherein: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method according to any one of claims 1 to 15.

20. An electronic device, wherein: include: a memory storing a computer program; A processor reads a computer program stored in a memory to execute the method according to any one of claims 1 to 15.

Images