Song background information processing method and device, equipment, medium and program product

Abstract

The invention provides a song background information processing method and device, equipment, a medium and a program product, and can relate to the AI technology, and the method comprises the steps: obtaining the basic information of a target song; wherein the basic information of the target song comprises a song name and a singer name of the target song; preliminarily selecting a real-time published article related to music from the at least one real-time published article based on the respective label of the at least one real-time published article; preliminarily selecting historical published articles related to the target song from the at least one historical published article based on the song name of the target song and the singer name; and obtaining background information of the target song based on the basic information of the target song, the preliminarily selected real-time published article related to the music, the preliminarily selected historical published article related to the target song and the large language model. Therefore, the generation efficiency and accuracy of the background information can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to a method, apparatus, device, medium and program product for processing background information of songs. Background Technology

[0002] Background information about a song can help listeners better understand it. An album, an artist, a playlist, or even a song may have its own unique creation story, legend, current status, and perspective for appreciation.

[0003] Currently, background information for songs is mainly generated manually, which relies on human experience and expertise, resulting in low efficiency and accuracy in generating background information. Summary of the Invention

[0004] This application provides a method, apparatus, device, medium, and program product for processing background information of songs, thereby improving the efficiency and accuracy of background information generation.

[0005] In a first aspect, embodiments of this application provide a method for processing background information of a song, comprising: obtaining basic information of a target song; wherein the basic information of the target song includes: the song title and the artist's name of the target song; based on the tags of at least one real-time published article, initially selecting real-time published articles related to music from at least one real-time published article; based on the song title and the artist's name of the target song, initially selecting historical published articles related to the target song from at least one historical published article; and obtaining background information of the target song based on the basic information of the target song, the initially selected real-time published articles related to music, the initially selected historical published articles related to the target song, and a large language model.

[0006] Secondly, embodiments of this application provide a method for processing background information of a song, comprising: obtaining a display operation for background information of a target song to generate a display request for background information of the target song; sending a display request to a backend server; receiving background information of the target song sent by the backend server in response to the display request; and displaying the background information of the target song; wherein the background information of the target song is obtained by the backend server in the following manner: based on the tags of at least one real-time published article, initially selecting real-time published articles related to music from at least one real-time published article; based on the song name and artist name of the target song, initially selecting historical published articles related to the target song from at least one historical published article; and obtaining the background information of the target song based on the basic information of the target song, the initially selected real-time published articles related to music, the initially selected historical published articles related to the target song, and a large language model.

[0007] Thirdly, embodiments of this application provide a song background information processing apparatus, including: an acquisition module, a selection module, and a generation module. The acquisition module acquires basic information about a target song, including the song title and artist name. The selection module initially selects music-related real-time published articles from at least one real-time published article based on their respective tags. The selection module also initially selects historically published articles related to the target song from at least one historically published article based on the song title and artist name. The generation module obtains the background information of the target song based on the basic information, the initially selected music-related real-time published articles, the initially selected historically published articles related to the target song, and a large language model.

[0008] In some implementations, the selection module is specifically used to: initially select music-related real-time published articles from at least one of the following: word count and author information of at least one real-time published article, and tags of at least one real-time published article.

[0009] In some implementations, the selection module is specifically used to: if the tag of the first real-time published article in at least one real-time published article is related to music, and the word count of the first real-time published article is greater than a first value, then the first real-time published article is selected as the initial selection of music-related real-time published articles.

[0010] In some implementations, the selection module is specifically used to: select the second real-time published article as the initial selection of music-related real-time published articles if the tag of the second real-time published article in at least one real-time published article is music-related, and the author information of the second real-time published article is not in the blacklist.

[0011] In some implementations, the selection module is specifically used to: select the third real-time published article as the initial selection of music-related real-time published articles if the tag of the third real-time published article is music-related, the author information of the third real-time published article is not in the blacklist, and the word count of the third real-time published article is greater than the second value.

[0012] In some implementations, the selection module is specifically used to: select the target historical article as a preliminary selection of historical articles related to the target song if at least one of the target historical articles contains the song title and artist name of the target song.

[0013] In some implementations, the generation module is specifically used to: extract keywords from the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song; match the song title and artist name of the target song with the keywords from the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song, in order to select articles with successful keyword matching; and obtain the background information of the target song based on the basic information of the target song, the articles with successful keyword matching, and the large language model.

[0014] In some implementations, the generation module is specifically used to: input the basic information of the target song and articles that have successfully matched the keywords into a large language model to obtain the background information of the target song.

[0015] In some implementations, the generation module is specifically used to: input the basic information of the target song, articles with successfully matched keywords, and prompt information into a large language model to obtain the background information of the target song.

[0016] In some implementations, the generation module is specifically used to: match the song title and artist name of the target song with the content of articles that successfully match the keywords, in order to select paragraphs with successfully matched content; and obtain the background information of the target song based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model.

[0017] In some implementations, the generation module is specifically used to: input the basic information of the target song and the paragraphs whose content is successfully matched into the large language model to obtain the background information of the target song.

[0018] In some implementations, the generation module is specifically used to: input the basic information of the target song, the paragraphs whose content is successfully matched, and the prompt information into the large language model to obtain the background information of the target song.

[0019] In some implementations, the device also includes a cleaning module for cleaning paragraphs whose content matches successfully.

[0020] In some implementations, the generation module is specifically used for: extracting features from a preliminary selection of real-time published articles related to music and a preliminary selection of historically published articles related to the target song based on a large language model; matching the song title, artist name, and extracted article features of the target song based on the large language model to select articles with successful feature matching; and processing the articles with successful feature matching based on the large language model to obtain background information of the target song.

[0021] In some implementations, the generation module is specifically used to: process articles and prompts that have successfully matched features based on a large language model to obtain the background information of the target song.

[0022] In some implementations, the prompt information includes at least one of the following: a summary of the task instructions, a definition of the background information, a prompt for the background information, format requirements for the background information, positive examples, and negative examples.

[0023] Fourthly, embodiments of this application provide a song background information processing apparatus, including: an acquisition module and a display module; wherein, the acquisition module is used to acquire display operations for comment information of a target song; the display module is used to display the comment information of the target song in a first display area; and display the background information of the target song in a second display area; wherein, the first display area and the second display area are different display areas; the background information of the target song is generated based on at least one article related to the target song.

[0024] In some possible implementations, the background information of the target song is generated as follows: based on the tags of at least one real-time published article, music-related real-time published articles are initially selected from at least one real-time published article; based on the song title and artist name of the target song, historical published articles related to the target song are initially selected from at least one historical published article; and based on the basic information of the target song, the initially selected music-related real-time published articles, the initially selected historical published articles related to the target song, and the large language model, the background information of the target song is obtained.

[0025] In some feasible methods, the initial selection of music-related live articles is obtained by: selecting music-related live articles from at least one live article based on at least one of the word count and author information of each live article, as well as the tags of each live article.

[0026] In some feasible methods, the initial selection of historical publications related to the target song is obtained as follows: if at least one historical publication contains the target historical publication and the artist's name, then the target historical publication is selected as the initial selection of historical publications related to the target song.

[0027] In some feasible implementations, the background information of the target song is obtained as follows: extracting keywords from a preliminary selection of music-related real-time published articles and a preliminary selection of music-related historical published articles; matching the song title and artist name of the target song with the keywords from the preliminary selection of music-related real-time published articles and the preliminary selection of music-related historical published articles to select articles with successful keyword matching; and obtaining the background information of the target song based on the basic information of the target song, the articles with successful keyword matching, and a large language model.

[0028] In some implementations, the device further includes a sending module, wherein the acquiring module is further configured to acquire feedback operations on the background information of the target song to generate feedback information on the background information of the target song; the sending module is configured to send the feedback information to a backend server so that the backend server adjusts the background information of the target song based on the feedback information.

[0029] In some possible implementations, feedback information includes: content errors, irrelevance, low quality, violation, or other reasons.

[0030] In some implementations, the device further includes: a receiving module, wherein the acquiring module is further configured to acquire a click operation targeting the "Understand Original Article" identifier to generate a request message for understanding the original article; the sending module is further configured to send a request message to a backend server to enable the backend server to retrieve the original article; the receiving module and the display module are further configured to receive and display the original article respectively.

[0031] Fifthly, embodiments of this application provide an electronic device, including: a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to perform methods as described in the first aspect, the second aspect, or their respective implementations.

[0032] Sixthly, embodiments of this application provide a computer-readable storage medium for storing a computer program that causes a computer to perform the methods described in the first aspect, the second aspect, or various implementations thereof.

[0033] In a seventh aspect, embodiments of this application provide a computer program product including computer program instructions that cause a computer to perform the methods as described in the first aspect, the second aspect, or various implementations thereof.

[0034] Eighthly, embodiments of this application provide a computer program that causes a computer to perform the methods described in the first aspect, the second aspect, or their respective implementations.

[0035] The technical solution provided in this application leverages the rich content of real-time and historically published articles, along with the powerful summarization and extraction capabilities of large language models, to automatically extract information from one or more articles or paragraphs within an article, and provide fluent background information for the song, thereby improving the efficiency and accuracy of background information generation. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of a system architecture according to an embodiment of this application;

[0038] Figure 2 A flowchart illustrating a method for processing song background information provided in this application embodiment;

[0039] Figure 3 A flowchart illustrating another method for processing song background information provided in this application embodiment;

[0040] Figures 4A to 4B This is a schematic diagram of the interface provided for an embodiment of this application;

[0041] Figures 5A to 5C This is a schematic diagram of the interface provided for an embodiment of this application;

[0042] Figures 6A to 6C This is a schematic diagram of the interface provided for an embodiment of this application;

[0043] Figure 7 A schematic diagram of a song background information processing device 700 provided in an embodiment of this application;

[0044] Figure 8 A schematic diagram of a song background information processing device 800 provided in an embodiment of this application;

[0045] Figure 9This is a schematic block diagram of the electronic device 900 provided in the embodiments of this application. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0048] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0049] The technical problems to be solved, the inventive concept and the system architecture of the embodiments of this application will be described below:

[0050] As mentioned above, the background information for songs is currently mainly generated manually, which relies on human experience and expertise, resulting in low efficiency and low accuracy in generating background information.

[0051] To address the aforementioned technical issues, this application proposes to leverage the rich content of real-time and historically published articles, while utilizing the powerful summarization and extraction capabilities of large language models, to automatically extract information from one or more articles or paragraphs within an article, and provide fluent background information for the song, thereby improving the efficiency and accuracy of background information generation.

[0052] In some possible implementations, the system architecture of embodiments of this application is as follows: Figure 1 As shown.

[0053] Figure 1 This is a schematic diagram of a system architecture according to an embodiment of this application. The system architecture involves network elements: terminal device 110 and backend server 120. The terminal device 110 and backend server 120 can be directly or indirectly connected through wired or wireless communication, which is not limited in this application.

[0054] It should be understood that the terminal device 110 may have a music application (APP) or an instant messaging client installed, wherein the instant messaging client provides a music interface to play songs and display background information of the songs through the music APP or music interface.

[0055] It should be understood that a music app is an application centered around music, allowing listeners to enjoy a range of functions such as listening to songs online, searching, saving, sharing, commenting, and creating music.

[0056] It should be understood that the instant messaging client can be any communication tool based on Internet technology that allows users to communicate with each other in real time through text, voice, video, etc., such as WeChat, QQ, WeChat Work, etc., but not limited to these.

[0057] It should be understood that the music interface is a built-in music and audio function of the instant messaging client, and can be regarded as a mini-program or functional module of the instant messaging client. Based on this, the music interface is also referred to as the music module or music mini-program in the instant messaging client. For example, the music interface could be WeChat's "Listen" feature, but it is not limited to this.

[0058] It should be understood that a mini program is a tiny application designed to perform a single function or a set of simple functions within a closed framework. They are typically embedded in larger applications or platforms, such as WeChat, and can be used directly without downloading or installing them.

[0059] It should be understood that WeChat Listen is an audio content service feature launched by the WeChat platform, designed to provide listeners with a convenient experience in accessing and listening to audio content.

[0060] In some possible implementations, the terminal device 110 may be a desktop computer, laptop computer, tablet computer, smartphone, smartwatch, virtual reality (VR), augmented reality (AR), etc., but is not limited to these.

[0061] In some implementations, the terminal device 110 may have a display module for displaying background information of the song, etc. The display module is also referred to as a monitor, display screen, touch screen, etc. The touch screen is also referred to as a touch screen, touch panel, etc.

[0062] It should be understood that backend server 120 refers to the backend server corresponding to the aforementioned music app or music interface, but is not limited to this.

[0063] In some implementations, the backend server 120 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0064] The backend server 120 can obtain basic information about the target song, including: the song title and artist name; based on the tags of at least one real-time published article, it can initially select music-related real-time published articles from at least one real-time published article; based on the song title and artist name of the target song, it can initially select historical published articles related to the target song from at least one historical published article; and based on the basic information of the target song, the initially selected music-related real-time published articles, the initially selected historical published articles related to the target song, and the large language model, it can obtain the background information of the target song.

[0065] In some possible implementations, the terminal device 110 can acquire the display operation of comment information for the target song; display the comment information of the target song in a first display area; and display the background information of the target song in a second display area; wherein the first display area and the second display area are different display areas; and the background information of the target song is generated based on at least one article related to the target song.

[0066] It should be noted that, Figure 1 This is merely a schematic diagram of a system architecture provided in this application embodiment; the system architecture involved in this application embodiment is not limited to... Figure 1 The system architecture shown, for example, Figure 1 The system architecture shown includes only one terminal device 110. In reality, it can include multiple terminal devices 110, or... Figure 1 Based on the system architecture shown, the system architecture involved in the embodiments of this application may also include other network elements, etc.

[0067] The technical solution of this application will be described in detail below:

[0068] Figure 2 This application provides a flowchart of a method for processing song background information. This method can be executed by a music app or a backend server corresponding to a music interface, for example, by... Figure 1 The backend server 120 shown is executed, but is not limited to this, such as... Figure 2 As shown, the method may include:

[0069] S210: Obtain basic information about the target song; whereby the basic information about the target song includes: the song title and the artist's name;

[0070] In some implementations, the target song can be any song for which background information is to be generated.

[0071] In some feasible implementations, the basic information of the target song may include at least one of the following, but is not limited to: the song title identifier (Identity, ID), the artist name ID, the year, genre, and lyrics of the target song.

[0072] It should be understood that a song title ID uniquely identifies a song. For example, if two songs have the same title but are actually two different songs, then their song title IDs will be different. Similarly, if two songs have the same title and content but different versions, then their song title IDs will also be different.

[0073] It should be understood that a singer's name ID can uniquely identify a singer. For example, if two singers have the same singer name but are actually two different singers, then their singer name IDs will be different.

[0074] In this embodiment of the application, the backend server may implement S210 in any of the following possible ways, but is not limited thereto:

[0075] In some implementations, S210 may include:

[0076] S210-1A: The backend server pre-stores the basic information of the target song in the local database;

[0077] S210-2A: The backend server retrieves basic information about the target song from the local database.

[0078] In some implementations, S210 may include:

[0079] S210-1B: The backend server obtains basic information about the target song from other music platforms through the Application Programming Interface (API).

[0080] In some implementations, S210 may include:

[0081] S210-1C: The backend server obtains basic information about the target song from multiple other music platforms via API interfaces;

[0082] S210-2C: The backend server integrates the basic information of the target song obtained from multiple other music platforms to obtain the final basic information of the target song.

[0083] For example, the backend server can obtain the song name and artist name of the target song from music platform 1, the song name ID and artist name ID of the target song from music platform 2, and the era, genre, and lyrics of the target song from music platform 3. Then, the backend server can combine the song name, artist name, song name ID, artist name ID, era, genre, and lyrics of the target song to form the basic information of the target song.

[0084] In some feasible implementations, the backend server can first obtain the song list, and then obtain the basic information of the target song in the song list.

[0085] In some possible implementations, the song list includes: a long-term popular song list and a short-term popular song list, or the song list includes: a male singer's song list and a female singer's song list, etc. In short, the embodiments of this application do not limit the way the song list is divided.

[0086] S220: Based on the tags of at least one real-time published article, initially select real-time published articles related to music from at least one real-time published article;

[0087] In some implementations, the at least one real-time published article can be a real-time published article on any article publishing platform, such as a real-time published article on the WeChat Official Accounts platform.

[0088] It should be understood that the tags for real-time published articles are used to distinguish the content type of real-time published articles. Based on this, the tags for real-time published articles are also called type tags or types of real-time published articles, etc., and this application embodiment does not limit this.

[0089] For example, the tags for articles published in real time include, but are not limited to: politics, emotions, life skills, lifestyle, technology, parenting, food, cars, knowledge, variety shows, anime, whether it is music, etc.

[0090] In this embodiment of the application, the backend server may implement S220 in any of the following possible ways, but is not limited thereto:

[0091] In some implementations, S220 may include:

[0092] S220-1A: Based on at least one of the word count and author information of at least one real-time published article, and the tags of at least one real-time published article, initially select real-time published articles related to music from at least one real-time published article.

[0093] In some possible implementations, based on at least one of the word count and author information of at least one real-time published article, and the tags of at least one real-time published article, a music-related real-time published article is initially selected from at least one real-time published article, including: if the tag of the first real-time published article among at least one real-time published article is music-related, and the word count of the first real-time published article is greater than a first value, then the first real-time published article is selected as the initial music-related real-time published article.

[0094] In some possible implementations, if at least one of the real-time published articles has a tag other than politics, emotions, life skills, life, technology, parenting, food, cars, or knowledge, then the backend server determines that the tag of that real-time published article is related to music.

[0095] In some implementations, if at least one of the live posts is tagged with "music", the backend server determines that the live post's tag is related to music.

[0096] It should be understood that this application does not impose any restrictions on how to determine whether the tags of real-time published articles are related to music.

[0097] It should be understood that actual tests have shown that large language models have difficulty obtaining useful background information from articles with fewer words. Instead, due to insufficient background information, they may produce various "illusions". Based on this, this application proposes to select articles with more than a first value for real-time publication.

[0098] In some possible implementations, the first value can be 100, 200, 300, etc., but is not limited to this.

[0099] For example, suppose a real-time published article is tagged with "music" and has 350 characters, which is greater than the first value of 300. Then the backend server can initially select this real-time published article.

[0100] In some possible implementations, based on at least one of the word count and author information of at least one real-time published article, and the tags of at least one real-time published article, a music-related real-time published article is initially selected from the at least one real-time published article, including: if the tag of a second real-time published article among the at least one real-time published article is music-related, and the author information of the second real-time published article is not in the blacklist, then the second real-time published article is selected as the initial music-related real-time published article.

[0101] It should be understood that, in the embodiments of this application, the blacklist refers to the list of authors of real-time published articles that cannot be used to generate song background information.

[0102] In some possible implementations, the blacklist includes, but is not limited to, author information from platforms or channels such as advertising, promotion, and concert information.

[0103] In some feasible implementations, author information includes, but is not limited to: author name, author gender, and author age.

[0104] For example, if a real-time published article is tagged with "music" and its author information is not on the blacklist, the backend server can initially select this real-time published article.

[0105] In some possible implementations, based on at least one of the word count and author information of at least one real-time published article, and the tags of at least one real-time published article, a music-related real-time published article is initially selected from the at least one real-time published article, including: if the tag of a third real-time published article among the at least one real-time published article is music-related, and the author information of the third real-time published article is not on the blacklist, and the word count of the third real-time published article is greater than the second value, then the third real-time published article is selected as the initial music-related real-time published article.

[0106] It should be understood that, regarding how to determine whether the tags of real-time published articles are related to music, please refer to the above, and this application embodiment will not repeat the details.

[0107] It should be understood that actual tests have shown that large language models have difficulty obtaining useful background information from articles with fewer words. Instead, due to insufficient background information, they may produce various "illusions". Based on this, this application proposes to select articles with more than a second value for real-time publication.

[0108] In some possible implementations, the second value can be 100, 200, 300, etc., but is not limited to this.

[0109] It should be understood that the explanation of the blacklist can be found above, and this application embodiment will not repeat it.

[0110] For example, suppose a real-time published article is tagged with "music" and has 350 characters, which is greater than the second value of 300, and its author information is not on the blacklist, then the backend server can initially select this real-time published article.

[0111] In some implementations, S220 may include:

[0112] S220-1B: If the fourth real-time post in at least one real-time post has a tag related to music, then the fourth real-time post is selected as the initial selection of music-related real-time posts.

[0113] For example, if a real-time published article is tagged with tags other than politics, emotions, life skills, lifestyle, technology, parenting, food, cars, knowledge, variety shows, and anime, then the backend server can initially select that real-time published article.

[0114] It should be understood that in S220, for real-time published articles, the backend server does not need to retrieve the content or main text of the real-time published articles. Compared to the subsequent need to retrieve the content or main text of real-time published articles, in S220, the backend server only needs to initially select real-time published articles based on their tags, or based on the tags and word count, or based on the tags and author information, or based on the tags, word count, and author information, for use in generating subsequent background information. Based on this, S220 can be referred to as a coarse matching or initial matching process between the target song and the real-time published articles.

[0115] S230: Based on the song title and artist name of the target song, initially select historical publications related to the target song from at least one historical publication;

[0116] In some implementations, the at least one historically published article can be a historically published article on any article publishing platform, such as a historically published article on the WeChat Official Accounts platform.

[0117] In this embodiment of the application, the backend server may implement S230 in any of the following ways, but is not limited thereto:

[0118] In some implementations, S230 may include:

[0119] S230-1A: If at least one historical publication contains the target historical publication title and artist name of the target song, then the target historical publication title is selected as the initial selection of historical publications related to the target song.

[0120] In some implementations, the backend server's local database stores each historically published article. When the backend server obtains the song title and artist name of the target song, it can extract features from each historically published article. Then, it matches the song title and artist name of the target song with the features of the historically published articles. If the features of a historically published article match both the song title and artist name of the target song, then the historically published article can be said to include the song title and artist name of the target song.

[0121] This feasible method can be understood as the backend server using a search engine to make an initial selection of historically published articles, where the song title and artist name of the target song are used as search terms.

[0122] In some implementation methods, the backend server performs feature extraction on each historically published article to extract features from the historically published articles. This includes: for any historically published article, the backend server can divide the historically published article into paragraphs and use a first neural network model to summarize the main content of each paragraph, and use the main content as the feature of the historically published article.

[0123] In some implementations, the first neural network model can be trained using training samples. Each training sample may include a text and the actual features of that text, where the actual features can serve as sample labels. The training device can employ supervised training; for example, it can input text into the first neural network model and output predicted features of that text. Furthermore, the training device can calculate a loss based on the actual and predicted features included in all training samples, and adjust the parameters of the first neural network model based on this loss until the training iterations reach a preset number or the loss reaches its minimum value, at which point training stops.

[0124] In some implementations, the training device may use any of the following loss functions when training the first neural network model, but is not limited to: L1 loss function, mean squared error (MSE) loss function, cross-entropy loss function, etc.

[0125] Based on this, the backend server can use the first neural network model to summarize the main content of each paragraph and use the main content as the feature of the historically published article.

[0126] In some implementation methods, the backend server divides historically published articles into paragraphs. This includes: the backend server can identify line breaks in historically published articles and divide the characters between two adjacent line breaks into a paragraph. For example, in a historically published article, there is a line break after the title and another line break after the first paragraph. The characters between these two line breaks are divided into a paragraph, namely the first paragraph. Furthermore, there is a line break after the second paragraph, and the characters between this line break and the line break at the end of the first paragraph are divided into another paragraph, namely the second paragraph, and so on. This completes the paragraph division of historically published articles.

[0127] In some implementations, the backend server divides historically published articles into paragraphs, including: the backend server can input historically published articles into a second neural network model to obtain paragraph division results for those historically published articles.

[0128] In some implementations, the second neural network model can be trained using training samples. Each training sample can include an article and the actual paragraph segmentation results of that article, where the actual paragraph segmentation results can serve as sample labels. The training device can employ supervised training methods; for example, it can input the article into the second neural network model and output the predicted paragraph segmentation results for that article. Furthermore, the training device can calculate a loss based on the actual and predicted paragraph segmentation results included in all training samples, and adjust the parameters of the second neural network model based on this loss until the training iterations reach a preset number or the loss reaches its minimum value, at which point training stops.

[0129] In some implementations, the training device may use any of the following loss functions when training the second neural network model, but is not limited to: L1 loss function, MSE loss function, cross-entropy loss function, etc.

[0130] Based on this, the backend server can use a second neural network model to divide historically published articles into paragraphs.

[0131] It should be noted that the embodiments of this application do not limit the paragraph division method.

[0132] In some implementations, the backend server performs feature extraction on each historically published article to extract features from the historically published articles. This includes: for any historically published article, the backend server can input the historically published article into a third neural network model to output the features of the historically published article.

[0133] In some implementations, the third neural network model can be trained using training samples. Each training sample can include an article and the article's actual features, where the actual features can serve as sample labels. The training device can employ supervised training; for example, it can input the article into the third neural network model and output the predicted features of the article. Furthermore, the training device can calculate a loss based on the actual and predicted features included in all training samples, and adjust the parameters of the third neural network model based on this loss until the training iterations reach a preset number or the loss reaches its minimum value, at which point training stops.

[0134] In some implementations, the training device may use any of the following loss functions when training the third neural network model, but is not limited to: L1 loss function, MSE loss function, cross-entropy loss function, etc.

[0135] Based on this, the backend server can use a third neural network model to extract features from historically published articles.

[0136] It should be noted that, for historically published articles, this application does not limit the method of feature extraction for the articles.

[0137] In some implementation methods, the backend server matches the song title and artist name of the target song with the features of historically published articles. This includes: the backend server calculating the similarity of each feature between the song title and the historically published articles, determining the maximum similarity corresponding to the song title; simultaneously calculating the similarity of each feature between the artist name and the historically published articles, determining the maximum similarity corresponding to the artist name; calculating the average of the maximum similarity corresponding to the song title and the maximum similarity corresponding to the artist name; if the average is greater than a first preset threshold, then the feature matching between the song title and artist name of the target song and the historically published articles is determined to be successful; if the average is less than or equal to the first preset threshold, then the feature matching between the song title and artist name of the target song and the historically published articles is determined to be unsuccessful.

[0138] In some implementation methods, the backend server matches the song title and artist name of the target song with the features of historically published articles. This includes: the backend server calculating the similarity of each feature between the song title and the historically published articles, determining the maximum similarity corresponding to the song title, and simultaneously calculating the similarity of each feature between the artist name and the historically published articles, determining the maximum similarity corresponding to the artist name, and calculating the maximum value of the maximum similarity corresponding to the song title and the maximum similarity corresponding to the artist name. If the maximum value is greater than a second preset threshold, it is determined that the features of the song title and artist name of the target song and the historically published articles have been successfully matched. If the maximum value is less than or equal to the second preset threshold, it is determined that the features of the song title and artist name of the target song and the historically published articles have failed to match.

[0139] In some implementation methods, the backend server matches the song title and artist name of the target song with the features of historically published articles. This includes: the backend server calculating the similarity of each feature between the song title and the historically published articles, determining the maximum similarity corresponding to the song title; simultaneously calculating the similarity of each feature between the artist name and the historically published articles, determining the maximum similarity corresponding to the artist name; calculating the minimum value between the maximum similarity corresponding to the song title and the maximum similarity corresponding to the artist name; if the minimum value is greater than a third preset threshold, then the feature matching between the song title and artist name of the target song and the historically published articles is determined to be successful; if the minimum value is less than or equal to the third preset threshold, then the feature matching between the song title and artist name of the target song and the historically published articles is determined to be unsuccessful.

[0140] In some implementations, the backend server calculates the similarity between the song title of the target song and each feature of the historically published articles. This includes: the backend server converting the artist name of the target song and each feature of the historically published articles into embeddings, and then calculating the similarity between the embeddings corresponding to the artist name and the embeddings corresponding to each feature of the historically published articles.

[0141] In some implementations, the backend server can use the Word2Vec model or the Bidirectional Encoder Representations from Transformers (BERT) model to convert each feature of the artist's name and historical published articles of the target song into an embedding vector, but is not limited to these.

[0142] In some implementations, when the backend server calculates the similarity between the embedding vector corresponding to the singer's name and the embedding vector corresponding to each feature of the historically published articles, it can use any of the following similarity calculation methods, but is not limited to: cosine similarity, Euclidean distance, Manhattan distance, Pearson correlation coefficient, Jaccard similarity coefficient, and Hamming distance.

[0143] In some implementations, the backend server's local database stores each historically published article. When the backend server obtains the song title and artist name of the target song, it can extract keywords from each historically published article. Then, it matches the song title and artist name of the target song with the keywords of the historically published articles. If the keywords of a historically published article match both the song title and artist name of the target song, then the historically published article can be said to include the song title and artist name of the target song.

[0144] This implementation method can also be understood as the backend server using a search engine to make an initial selection of historically published articles, where the song title and artist name of the target song are used as search terms.

[0145] In some possible implementations, the backend server extracts keywords for each historically published article. This includes: for each historically published article, the backend server can first divide the article into words, then calculate the term frequency-inverse document frequency (TF-IDF) of each word, and use words with TF-IDF greater than a preset frequency as keywords in the article.

[0146] In some feasible implementations, the backend server can use dictionary-based word segmentation methods, statistical model-based word segmentation methods, or deep learning-based word segmentation methods to segment words in the article, but is not limited to these.

[0147] It should be understood that dictionary-based word segmentation methods mainly utilize pre-built dictionaries to match and segment articles according to the words in the dictionary. If the character sequence in the article matches the words in the dictionary exactly, it is segmented into the corresponding words; otherwise, it continues to match according to certain rules or processes out-of-vocabulary words. Common matching methods include forward maximum matching, backward maximum matching, and bidirectional maximum matching.

[0148] It should be understood that word segmentation methods based on statistical models mainly segment words by statistically analyzing the frequency of word occurrences and collocation relationships in a large corpus. Common algorithms include Hidden Markov Models (HMMs) and Conditional Random Fields (CRFs), which can predict the most likely word segmentation positions based on contextual information.

[0149] It should be understood that deep learning-based word segmentation methods use neural network models to annotate each character or word, thereby achieving word segmentation.

[0150] In some implementations, the backend server extracts keywords for each historically published article, including: for each historically published article, the backend server can input the article into a fourth neural network model to output the keywords of the article.

[0151] In some implementations, the fourth neural network model can be trained using training samples. Each training sample may include an article and the article's actual keywords, where the actual keywords can serve as sample labels. The training device can employ supervised training; for example, it can input the article into the fourth neural network model and output predicted keywords for the article. Furthermore, the training device can calculate a loss based on the actual and predicted keywords included in all training samples, and adjust the parameters of the fourth neural network model based on this loss until the training iterations reach a preset number or the loss reaches its minimum value, at which point training stops.

[0152] In some implementations, the training device may use any of the following loss functions when training the fourth neural network model, but is not limited to: L1 loss function, MSE loss function, cross-entropy loss function, etc.

[0153] Based on this, the backend server can use the fourth neural network model to extract keywords from historically published articles.

[0154] It should be noted that this application does not restrict the keyword extraction method for previously published articles.

[0155] In some implementations, S230 may include:

[0156] S230-1B: If at least one of the target historical publications includes the song title or artist name of the target song, then the target historical publication is selected as the initial selection of historical publications related to the target song.

[0157] In some implementations, the backend server's local database stores each historically published article. When the backend server obtains the song title and artist name of the target song, it can perform feature extraction on each historically published article to extract features from the historically published articles. Then, it matches the song title of the target song with the features in the historically published articles, or it matches the artist name of the target song with the features in the historically published articles. If a historically published article matches the song title or artist name of the target song, then the historically published article can be said to include the song title or artist name of the target song.

[0158] In some implementations, the backend server's local database stores each historically published article. When the backend server obtains the song title and artist name of the target song, it can extract keywords from each historically published article. Then, it matches the target song title with the keywords in the historically published articles, or it matches the target song artist name with the keywords in the historically published articles. If a historically published article matches the target song title or artist name, then the historically published article can be said to contain the target song title or artist name.

[0159] It should be understood that the methods for feature extraction, keyword extraction, feature matching, and keyword matching can be referred to above, and will not be repeated in the embodiments of this application.

[0160] It should be noted that the difference between this implementation method and the previous implementation method of S230 is that in this implementation method, the historical published articles only need to match the song name or singer name of the target song, while in the previous implementation method, the historical published articles need to match both the song name and singer name of the target song. Based on this, this implementation method obtains more historical published articles, but the accuracy is lower, while the previous implementation method obtains relatively fewer historical published articles, but the accuracy is higher.

[0161] It should be understood that in S230, for historically published articles, compared to the content or main text of the historically published articles that need to be retrieved later, S230 can be referred to as a coarse matching or initial matching process between the target song and the historically published articles.

[0162] S240: Based on the basic information of the target song, the preliminary selection of real-time published articles related to music, the preliminary selection of historical published articles related to the target song, and the large language model, the background information of the target song is obtained.

[0163] It should be understood that background information of a song refers to a series of information related to the creation, recording, release, and related cultural and historical context of the target song. This information helps listeners to understand the song's intentions, emotions, and its significance in a specific period or cultural context more deeply.

[0164] For example, one of the background information for Song 1 is: "...because its lyrics convey positive energy and have a subtle educational significance, it was selected as a first-grade textbook for the Affiliated Experimental Primary School of Education University...".

[0165] For example, one of the background details for song 2 is: "This song was written specifically for college students leaving school. Singer A voluntarily told his story to the songwriter, and thus this graduation song was born. When recording this song, a home studio recording method was used. The recording location was the dining room of Singer A's home. An untuned piano was deliberately used, and a regular home piano was specially chosen instead of a grand piano for performance. Instead of advanced and fast digital recording, traditional analog master tape recording technology was used."

[0166] It should be understood that a Large Language Model (LLM) is an artificial intelligence model based on deep learning technology that uses a large amount of text data to understand and generate natural language. In the embodiments of this application, the powerful summarization and extraction capabilities of the Large Language Model are mainly utilized to automatically extract information from one or more articles or paragraphs within an article, and provide fluent background information for the song.

[0167] In this embodiment of the application, the backend server may implement S240 in any of the following possible ways, but is not limited thereto:

[0168] In some implementations, S240 may include:

[0169] S240-1A: Extract keywords from both the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song;

[0170] In some feasible implementations, the backend server can extract keywords from the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song in real time.

[0171] In some feasible implementations, the backend server can periodically extract keywords from the initially selected real-time articles related to music and the initially selected historical articles related to the target song after they have been published.

[0172] It should be understood that, for the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song, this application embodiment does not limit the timing of keyword extraction.

[0173] It should be understood that the keyword extraction method can be referred to in S230, and will not be repeated in this embodiment.

[0174] It should be understood that if S230 extracts keywords from historically published articles, then the keyword extraction methods for historically published articles involved in S230 and S240-1A can be different. For example, one may extract keywords based on TF-IDF, while the other may extract keywords based on a neural network model.

[0175] S240-2A: Match the song title and artist name of the target song with the keywords of the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song, and select the articles with successful keyword matching;

[0176] In some implementation methods, the backend server matches the song title and artist name of the target song with the keywords of the pre-selected real-time and historically published music-related articles. This includes: for any one of the pre-selected real-time and historically published music-related articles, the backend server can calculate the similarity between the song title and each keyword of the article, determine the maximum similarity corresponding to the song title, and simultaneously calculate the similarity between the artist name and each keyword of the article, determine the maximum similarity corresponding to the artist name, and calculate the average of the maximum similarity corresponding to the song title and the maximum similarity corresponding to the artist name. If the average is greater than a fourth preset threshold, it is determined that the song title and artist name of the target song and the keywords of the article have successfully matched; if the average is less than or equal to the fourth preset threshold, it is determined that the song title and artist name of the target song and the keywords of the article have failed to match.

[0177] In some implementation methods, the backend server matches the song title and artist name of the target song with the keywords of the pre-selected real-time and historically published music-related articles. This includes: for any one of the pre-selected real-time and historically published music-related articles, the backend server calculates the similarity between the song title and each keyword of the article, determines the maximum similarity corresponding to the song title, and simultaneously calculates the similarity between the artist name and each keyword of the article, determines the maximum similarity corresponding to the artist name, and calculates the maximum value of the maximum similarity corresponding to the song title and the maximum similarity corresponding to the artist name. If the maximum value is greater than a fifth preset threshold, the song title and artist name of the target song are considered to have successfully matched the keywords of the article; if the maximum value is less than or equal to the fifth preset threshold, the song title and artist name of the target song are considered to have failed to match the keywords of the article.

[0178] In some implementation methods, the backend server matches the song title and artist name of the target song with the keywords of the pre-selected real-time and historically published music-related articles. This includes: for any one of the pre-selected real-time and historically published music-related articles, the backend server calculates the similarity between the song title and each keyword of the article, determining the maximum similarity for the song title; simultaneously, it calculates the similarity between the artist name and each keyword of the article, determining the maximum similarity for the artist name; and calculates the minimum value between the maximum similarity for the song title and the maximum similarity for the artist name. If this minimum value is greater than a sixth preset threshold, the match between the song title and artist name and the keywords of the article is considered successful. If the minimum value is less than or equal to the sixth preset threshold, the match between the song title and artist name and the keywords of the article is considered unsuccessful.

[0179] It should be understood that the similarity calculation method can be referred to in S230, and will not be repeated in this embodiment.

[0180] S240-3A: Based on the basic information of the target song, articles with successfully matched keywords, and a large language model, obtain the background information of the target song.

[0181] In the embodiments of this application, S240-3A can be implemented in any of the following possible ways, but is not limited thereto:

[0182] Option 1 can be implemented, and S240-3A may include:

[0183] S240-3A-1a: Input the basic information of the target song and the articles that successfully match the keywords into the large language model to obtain the background information of the target song.

[0184] In some implementations, multiple articles can be successfully matched with keywords. Based on this, for each article that is successfully matched with keywords, the backend server can use a large language model to generate background information for the target song and the corresponding score for that background information. The higher the score, the higher the accuracy of the background information; conversely, the lower the score, the lower the accuracy of the background information.

[0185] It should be understood that if the basic information of the target song includes its song name ID, and since the song name ID can uniquely identify a song (i.e., it can distinguish between songs with the same name but different content, or songs with the same name and content but different versions), and the backend server has been trained to learn the meaning represented by the song name ID, then the backend server can generate more accurate background information based on the song name ID of the target song. For example, for song A, its song name is A, but it corresponds to two different songs with song name IDs 1 and 2 respectively. Even if the backend server generates background information with IDs 1 and 2 using the same article, the generated background information will be the same, but the corresponding scores will be different. For the backend server, for the song corresponding to the same song name ID, it can generate different background information based on different articles, and ultimately select the background information with the highest score. This ensures that two songs with the same name but different song name IDs have different background information.

[0186] It should be understood that if the basic information of the target song includes the singer's name ID, and since the singer's name ID can uniquely identify a singer (i.e., distinguish different singers), and the backend server has been trained to learn the meaning represented by the singer's name ID, then the backend server can generate more accurate background information based on the singer's name ID of the target song. For example, for a singer named Zhang San, whose singer name is Zhang San, but who corresponds to two different singers with singer name IDs 01 and 02 respectively, even if the backend server uses the same article to generate background information with IDs 01 and 02, the generated background information is the same, but the corresponding scores are different. For the backend server, for the same singer with the same singer name ID, it can generate different background information based on different articles, and finally select the background information with the highest score. Thus, two singers with the same singer name but different singer name IDs will have different background information for the songs they sing.

[0187] In some implementations, before inputting the basic information of the target song and the articles with successfully matched keywords into the large language model, the backend server can also clean the articles with successfully matched keywords. This includes filtering out at least one of the following invalid information from the articles with successfully matched keywords: text with links, hidden text, garbled characters, playlists with no informational content, and lyrics. Since lyrics are an attribute of the song, they can interfere with the subsequent generation of background information; therefore, the backend server can filter out lyrics from articles with successfully matched keywords.

[0188] It should be understood that cleaning articles that successfully match keywords can filter out invalid information, thereby improving the accuracy and efficiency of background information generation.

[0189] Option 2 can be implemented, and S240-3A may include:

[0190] S240-3A-1b: Input the basic information of the target song, the articles with successfully matched keywords, and the prompt information into the large language model to obtain the background information of the target song.

[0191] It should be understood that the beneficial effects and examples of introducing song name ID and artist name ID can be found in the explanation of S240-3A-1a, and will not be repeated in this embodiment.

[0192] It should be understood that prompt engineering plays an indispensable role in large language models. Prompts, as input to the large language model, directly affect its output. Through prompt engineering, the model's performance can be enhanced without changing its parameters. By modifying the prompts, the large language model can perform different downstream tasks. Purposeful and directional adjustments to the prompts can further improve the output performance of the large language model.

[0193] In prompt engineering, a commonly used method is called Chain-of-Thought (CoT). As an improvement technique for prompts, CoT primarily aims to enhance the performance of large language models in reasoning about complex problems. The main idea of ​​CoT is to enhance the capabilities of large language models by explicitly outputting a portion of the reasoning steps and thought processes, mimicking human thought processes, before outputting the final result.

[0194] In this application embodiment, a CoT-compliant prompt message is introduced.

[0195] In some implementations, the prompt information includes at least one of the following: a summary of the task instructions, a definition of the background information, a prompt for the background information, format requirements for the background information, positive examples, and negative examples.

[0196] It should be understood that a copying task instruction is used to instruct on a copying task, which involves extracting content from an article or paragraph. For example, a copying task instruction might be: "You are a music expert; you know all the songs, musicians, albums, and various music-related technical terms and knowledge. You answer the questions using precise language without producing redundant output. Based on the given literature, introduce the musical piece."

[0197] You will be tasked with taking notes.

[0198] It should be understood that the definition of background information is used to define the type of background information. For example, the definition of background information includes the following: 1. The background of the song's creation; 2. The content of the song; 3. The feelings and insights that the song evokes; 4. The characteristics of the song.

[0199] It should be understood that background information prompts are used to indicate precautions when generating background information. For example, a background information prompt might be: "I will give you an article about a certain piece of music. This article describes a specific piece of music, and you only need to focus on the content related to this piece. Remove any text in the article that is irrelevant to the music or meaningless. Ignore the lyrics in the article. Extract the background information for the specified piece of music from the article."

[0200] The following is an example illustrating the formatting requirements for background information:

[0201] For example, the format requirements for the background information are: "The excerpt should be between 100 and 200 words. The output format is a single-line JSON. The JSON has three fields: {{"Song Name":"","Excerpt Content":"","Whether it meets the background information requirements":""}}.

[0202] It should be understood that positive examples refer to positive instances of background information, also known as positive samples of background information. Similarly, negative examples refer to negative instances of background information, also known as negative samples of background information.

[0203] For example, the background information is as follows: "The article title is <Song A | Time Rewind, Are You Ready?>, the article content is <Let's Talk About the Magical Song 'Song A', which is included in Zhang San's Super Popular Debut Album 1...>, extract the content describing Song A from the article. The output is: {{"Song Name":""Song A",Excerpt Content":"The magical song 'Song A' is included in Zhang San's Super Popular Debut Album 1, ranking tenth, but it is definitely a low-key powerhouse. Most of the songs on this album have a touch of Western style, but this one is different. It is purely Chinese style, giving you a surprise right from the start, playing a little trick with Chinese pinyin, which sounds as familiar and fresh as learning pinyin as a child. The whole song is simple but not simple, like masters sparring, without too many fancy tricks, just three chords, yet it manages to create a deep and beautiful oriental sentiment.",Does it meet the requirements of the music story":"Yes"}}.

[0204] For example, a negative example of background information is as follows: "The original text is <Today I'm sharing a new hot song from the charts. It's from 'Zhang San - Song B'. It's super catchy...>. This article cannot be excerpted because it does not contain descriptions or evaluations of the music. The beginning lists lyrics and basic song information, not descriptions of the music's characteristics or background information. Most of the later content is unrelated to the music and cannot be excerpted. The unqualified output is {{"Song Name":"Song B","Excerpt Content":"The article does not describe the song's content","Does it meet the requirements of a music story":"No"}}. The output is unqualified because it contains redundant explanatory statements and is not a direct excerpt of the original text."

[0205] The prompts clearly define the behavior of the large language model, and by strictly formatting the background information, the prediction process is simplified. Furthermore, following the principles of CoT (Coding on Tolerance), even if the song title is given in the prompt, the large language model still needs to output the song title. Additionally, by adding a decision field to the output, requiring the large language model to further determine whether the output matches the prompt, the task of initially judging the output by calling the large language model twice—one generation task and one decision task—is optimized, significantly reducing the prediction difficulty.

[0206] In some implementations, the backend server can clean the articles with successfully matched keywords before inputting the basic information of the target song, the articles with successfully matched keywords, and the prompt information into the large language model. The cleaning method and its corresponding beneficial effects can be found above, and will not be elaborated further in this embodiment.

[0207] Option 3 can be implemented, and S240-3A may include:

[0208] S240-3A-1c: Match the content of articles in which the song title and artist name of the target song are successfully matched with the keywords, and select the paragraphs in which the content is successfully matched;

[0209] It should be understood that the backend server can match the song title and artist name of the target song with each paragraph in the article that successfully matches the keywords, in order to select the paragraphs whose content matches successfully.

[0210] It should be understood that the paragraph division method of the article can be referred to above, and this application embodiment will not repeat it.

[0211] In some feasible methods, for each paragraph in an article where keywords are successfully matched, the backend server can perform word segmentation on that paragraph and match each word with the song title and artist name of the target song. If a paragraph matches both the song title and artist name of the target song, then that paragraph can be considered a paragraph with successful content matching.

[0212] It should be understood that the vocabulary classification method can be referred to above, and will not be repeated in this application embodiment.

[0213] It should be understood that the matching method for words and the song title and artist name of the target song can be referred to above, and will not be repeated in this embodiment.

[0214] In some implementations, after the backend server obtains the paragraphs that have successfully matched the content, it can establish a correspondence between the target song and the paragraphs that have successfully matched the content, and format the correspondence into a file required by the large language, such as a JSON file, and save it to the storage medium so that the large language model can pull the file and generate the background information of the target song.

[0215] S240-3A-2c: Based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model, obtain the background information of the target song.

[0216] In some possible implementations, the backend server obtains the background information of the target song based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model. This includes: the backend server inputting the basic information of the target song and the paragraphs with successfully matched content into the large language model to obtain the background information of the target song.

[0217] It should be understood that the beneficial effects and examples of introducing song name ID and artist name ID can be found in the explanation of S240-3A-1a, and will not be repeated in this embodiment.

[0218] In some implementations, before inputting the basic information of the target song and the successfully matched paragraphs into the large language model, the backend server can also clean the successfully matched paragraphs. This includes filtering out at least one of the following invalid information from the successfully matched paragraphs: text with links, hidden text, garbled characters, playlists with no informational content, and lyrics. Since lyrics are an attribute of the song and can interfere with the subsequent generation of background information, the backend server can filter out lyrics from the successfully matched paragraphs.

[0219] It should be understood that cleaning paragraphs that successfully match the content can filter out invalid information, thereby improving the accuracy and generation efficiency of background information.

[0220] In some possible implementations, the backend server obtains the background information of the target song based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model. This includes: the backend server inputting the basic information of the target song, the paragraphs with successfully matched content, and the prompt information into the large language model to obtain the background information of the target song.

[0221] It should be understood that the explanation of the prompt information can be found above, and this application embodiment will not repeat it.

[0222] In some implementations, the backend server can clean the successfully matched segments before inputting the basic information of the target song, the segments with successfully matched content, and the prompt information into the large language model. The cleaning method and its corresponding beneficial effects can be found above, and will not be elaborated further in this embodiment.

[0223] It should be noted that the implementation methods comprised of S240-1A to S240-3A are actually a fine-grained matching or re-matching process between the target song and the article. For the implementation method provided by S240-3A, the backend server can locate the paragraphs where the content is successfully matched, so that the subsequent large language model can generate background information, or in other words, extract background information.

[0224] In some implementations, S240 may include:

[0225] S240-1B: Based on a large language model, feature extraction is performed on a preliminary selection of real-time published articles related to music and a preliminary selection of historical published articles related to the target song;

[0226] It should be understood that the feature extraction method can be referred to above, and the embodiments of this application will not be repeated here.

[0227] S240-2B: Based on a large language model, the song title, artist name, and extracted article features of the target song are matched to select articles with successful feature matching;

[0228] It should be understood that the backend server needs to simultaneously match the song title of the target song with the extracted article features, as well as the singer's name with the extracted article features, in order to select the article with a successful feature match. The feature matching method can be referred to above, and will not be repeated here in this embodiment.

[0229] S240-3B: Based on a large language model, the article with successfully matched features is processed to obtain the background information of the target song.

[0230] In some feasible implementations, the article with successful feature matching can be a single article. Based on this, the backend server can use a large language model to locate paragraphs in the article with successful feature matching that contain the song title and artist name of the target song, and extract these paragraphs to obtain the background information of the target song.

[0231] In some feasible implementations, there can be multiple articles with successful feature matching. Based on this, for each article with successful feature matching, the backend server can use a large language model to locate the paragraphs in the article that contain the song name and artist name of the target song, extract these paragraphs, and merge all the extracted content to obtain the background information of the target song.

[0232] In some implementations, the backend server can process the articles and prompts that successfully match the features based on a large language model to obtain the background information of the target song. The articles that successfully match the features can be one or more.

[0233] It should be understood that the explanation of the prompt information can be found above, and this application embodiment will not repeat it.

[0234] It should be noted that, compared to the previous implementation method of S240, this implementation method can be understood as integrating the fine matching process of the target song and the article into the large language model, forming an end-to-end model strategy.

[0235] In this implementation, the large language model can be a Retrieval-Augmented Generation (RAG) model. The core idea of ​​a RAG model is to retrieve relevant knowledge information before generating text, and then combine this information with the input information to generate a more accurate and useful answer or text. In this embodiment, the articles obtained through coarse matching can be stored in a knowledge base for retrieval by the RAG model, thus omitting the fine matching process between the target song and the articles. Furthermore, in this implementation, prompts can be used to request the large language model to provide relevant articles about the song and extract summaries.

[0236] This application provides a method for processing background information of a song, comprising: a backend server obtaining basic information of a target song; wherein the basic information of the target song includes: the song title and the artist's name; based on the tags of at least one real-time published article, initially selecting music-related real-time published articles from at least one real-time published article; based on the song title and the artist's name of the target song, initially selecting historically published articles related to the target song from at least one historically published article; and obtaining background information of the target song based on the basic information of the target song, the initially selected music-related real-time published articles, the initially selected historically published articles related to the target song, and a large language model. By leveraging the rich content of real-time and historically published articles, and utilizing the powerful summarization and extraction capabilities of the large language model, the method automatically extracts information from one or more articles or paragraphs within articles, providing grammatically correct background information for the song, thereby improving the efficiency and accuracy of background information generation.

[0237] Furthermore, since real-time and historical articles are dynamically updated, the background information of the songs generated based on real-time and historical articles is also dynamically updated, which can improve the listener's experience.

[0238] Furthermore, this application leverages the powerful summarization and extraction capabilities of large language models to extract essential summaries from one or more articles, enabling listeners to quickly understand the background information of a song without having to jump to the original text, thereby enhancing the listener experience.

[0239] Furthermore, current technologies only provide background information at the level of "album," "playlist," or "artist," failing to provide a comprehensive understanding of each song. Individual songs often possess unique background information, increasing their information content. For example, one piece of background information for song 1 is: "...because its lyrics convey positive energy and have subtle educational significance, it was selected for the first-grade textbook of the Affiliated Experimental Primary School of University of Education A...". The method for generating song background information provided in this application can reach the song level, which is more granular than "album," "playlist," or "artist," thus enhancing the listener experience.

[0240] In addition, from another perspective, background information about a song can also serve to connect the song and the article, helping to increase the article's exposure and dissemination.

[0241] It should be noted that, in order to improve the prediction ability of the large language model, this application embodiment also proposes to enhance the training of the large language model and fine-tune the prompt information. Furthermore, the training samples of the large language model can also be optimized, for example, by discarding training samples with too short background information, discarding training samples lacking background information, discarding training samples with incorrect background information and song matching, etc.

[0242] It is important to note that even after retraining the large language model or fine-tuning the prompts, it is still necessary to monitor the state and behavior of the large language model during the prediction process in order to dynamically update and adjust the large language model and prompts.

[0243] Figure 3 This is a flowchart illustrating another method for processing song background information provided in an embodiment of this application. This method can be executed by a terminal device. The terminal device has a music app or an instant messaging client installed. The instant messaging client provides a music interface to play songs and display their background information through the music app or interface. The terminal device can be... Figure 1 The terminal device 110 shown is not limited to this, such as... Figure 3 As shown, the method may include:

[0244] S310: Terminal device retrieves and displays comment information for the target song;

[0245] It should be understood that the operation of displaying comment information for a target song refers to the operation used to trigger the display of comment information for the target song.

[0246] In some implementations, the display operation can be a click on the comment icon of the target song, or a double-click on the playback interface of the target song, but is not limited to these.

[0247] It should be understood that after receiving the display operation, the terminal device can respond to it by generating a display request. This display request is used to request the backend server to send the comment information and background information of the target song, in order to display the comment information and background information of the target song.

[0248] After receiving the display request, the backend server can send the comment information and background information of the target song to the terminal device.

[0249] S320: The terminal device displays comment information for the target song in the first display area;

[0250] S330: The terminal device displays the background information of the target song in the second display area; wherein the first display area and the second display area are different display areas; the background information of the target song is generated based on at least one article related to the target song.

[0251] For example, Figures 4A to 4B This is a schematic diagram of the interface provided for an embodiment of this application, such as... Figure 4A As shown, when listeners click the comment icon for "Song 1", they enter a menu similar to... Figure 4B The interface shown displays background information and comments for "Song 1", which are located in different display areas.

[0252] In some possible implementations, the background information of the target song is generated as follows: based on the tags of at least one real-time published article, music-related real-time published articles are initially selected from at least one real-time published article; based on the song title and artist name of the target song, historical published articles related to the target song are initially selected from at least one historical published article; and based on the basic information of the target song, the initially selected music-related real-time published articles, the initially selected historical published articles related to the target song, and the large language model, the background information of the target song is obtained.

[0253] It should be understood that the specific process for generating the background information of the target song can be referenced. Figure 2 The corresponding embodiments will not be described in detail in this application.

[0254] In some implementations, the song background information processing method further includes: a terminal device acquiring feedback operations on the background information of a target song to generate feedback information on the background information of the target song; and sending the feedback information to a backend server so that the backend server adjusts the background information of the target song based on the feedback information.

[0255] It should be understood that the feedback operation for the background information of the target song refers to the operation used to provide feedback on the background information of the target song.

[0256] In some possible implementations, the feedback operation includes: clicking on the source information of the background information of the target song, clicking on the feedback identifier, and selecting the feedback information. This application embodiment does not limit this.

[0257] It should be understood that the feedback information includes, but is not limited to: content errors, irrelevance, low quality, violation of regulations, and other reasons.

[0258] For example, Figures 5A to 5C This is a schematic diagram of the interface provided for an embodiment of this application, such as... Figure 5A As shown, listeners can click on the source information in the background information of "Song 1" to enter... Figure 5B The interface shown allows listeners to click the "Feedback" icon to access... Figure 5C The interface shown displays feedback information, including: content error, content irrelevant, content low quality, content violation, and other reasons.

[0259] In some implementations, the song background information processing method further includes: the terminal device receiving a click operation on the "Learn the original article" identifier to generate a request message for learning the original article; and sending the request message to the backend server so that the backend server can retrieve the original article and send the original article to the terminal device so that the terminal device can display the original article.

[0260] It should be understood that the original article refers to the source article of the background information, that is, the background information is excerpted from that article.

[0261] For example, Figures 6A to 6C This is a schematic diagram of the interface provided for an embodiment of this application, such as... Figure 6A As shown, listeners can click on the source information in the background information of "Song 1" to enter... Figure 6B The interface shown allows listeners to click the "Read the original article" icon to access the page. Figure 6C The interface shown displays the original article content.

[0262] In this embodiment, listeners can provide feedback on the background information of a song, thereby optimizing and improving the background information and increasing its accuracy. Furthermore, listeners can also access the original article from which the background information originates, which enhances the listener experience and helps increase the article's exposure and dissemination.

[0263] The preferred embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this application, various simple modifications can be made to the technical solutions of this application, and these simple modifications all fall within the protection scope of this application. For example, the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this application will not describe the various possible combinations separately. Furthermore, various different embodiments of this application can also be arbitrarily combined, as long as they do not violate the spirit of this application, they should also be considered as the content disclosed in this application.

[0264] It should also be understood that, in the various method embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0265] The method provided in the embodiments of this application has been described above. The song background information processing device provided in the embodiments of this application will be described below.

[0266] Figure 7 A schematic diagram of a song background information processing device 700 provided in an embodiment of this application is shown below. Figure 7 As shown, the device 700 includes: an acquisition module 710, a selection module 720, and a generation module 730.

[0267] The acquisition module 710 is used to acquire the basic information of the target song; wherein, the basic information of the target song includes: the song name and the artist name of the target song;

[0268] The selection module 720 is used to initially select music-related live-posted articles from at least one live-posted article based on the tags of each of the at least one live-posted article.

[0269] The selection module 720 is also used to initially select historically published articles related to the target song from at least one historically published article based on the song name and artist name of the target song;

[0270] The generation module 730 is used to obtain the background information of the target song based on the basic information of the target song, the initially selected real-time published articles related to music, the initially selected historical published articles related to the target song, and the large language model.

[0271] In some implementations, the selection module 720 is specifically used to: initially select music-related real-time published articles from at least one of the following: word count and author information of each of the at least one real-time published articles, and the tags of each of the at least one real-time published articles.

[0272] In some implementations, the selection module 720 is specifically used to: if the tag of the first real-time published article in at least one real-time published article is related to music, and the word count of the first real-time published article is greater than a first value, then the first real-time published article is selected as the initial selection of music-related real-time published articles.

[0273] In some implementations, the selection module 720 is specifically used to: select the second real-time published article as a preliminary selection of music-related real-time published articles if the tag of the second real-time published article in at least one real-time published article is music-related and the author information of the second real-time published article is not in the blacklist.

[0274] In some implementations, the selection module 720 is specifically used to: select the third real-time published article as the initial selection of music-related real-time published articles if the tag of the third real-time published article is music-related, the author information of the third real-time published article is not in the blacklist, and the word count of the third real-time published article is greater than the second value.

[0275] In some implementations, the selection module 720 is specifically used to: select the target historical publication article as a preliminary selection of historical publication articles related to the target song if at least one historical publication article contains the song title and artist name of the target song.

[0276] In some implementations, the generation module 730 is specifically used to: extract keywords from the pre-selected real-time published articles related to music and the pre-selected historical published articles related to the target song; match the song title and artist name of the target song with the keywords from the pre-selected real-time published articles related to music and the pre-selected historical published articles related to the target song to select articles with successful keyword matching; and obtain the background information of the target song based on the basic information of the target song, the articles with successful keyword matching, and the large language model.

[0277] In some implementations, the generation module 730 is specifically used to: input the basic information of the target song and the articles that have successfully matched the keywords into the large language model to obtain the background information of the target song.

[0278] In some implementations, the generation module 730 is specifically used to: input the basic information of the target song, the articles with successfully matched keywords, and the prompt information into the large language model to obtain the background information of the target song.

[0279] In some implementations, the generation module 730 is specifically used to: match the song title and artist name of the target song with the content of articles that successfully match the keywords, in order to select paragraphs with successfully matched content; and obtain the background information of the target song based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model.

[0280] In some implementations, the generation module 730 is specifically used to: input the basic information of the target song and the paragraphs whose content is successfully matched into the large language model to obtain the background information of the target song.

[0281] In some implementations, the generation module 730 is specifically used to: input the basic information of the target song, the paragraphs whose content is successfully matched, and the prompt information into the large language model to obtain the background information of the target song.

[0282] In some implementations, the device 700 also includes a cleaning module 740 for cleaning paragraphs whose content is successfully matched.

[0283] In some implementations, the generation module 730 is specifically used to: extract features from a preliminary selection of music-related real-time published articles and a preliminary selection of historically published articles related to the target song based on a large language model; match the song title, artist name, and extracted article features of the target song based on the large language model to select articles with successful feature matching; and process the articles with successful feature matching based on the large language model to obtain background information of the target song.

[0284] In some implementations, the generation module 730 is specifically used to: process the articles and prompts that have successfully matched features based on a large language model to obtain the background information of the target song.

[0285] In some implementations, the prompt information includes at least one of the following: a summary of the task instructions, a definition of the background information, a prompt for the background information, format requirements for the background information, positive examples, and negative examples.

[0286] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 7 The device 700 shown can perform Figure 2 The corresponding method embodiments are provided, and the foregoing and other operations and / or functions of each module in the device 700 are respectively for implementing... Figure 2For the sake of brevity, the corresponding processes in each method are not described in detail here.

[0287] The apparatus 700 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.

[0288] Figure 8 A schematic diagram of a song background information processing device 800 provided in an embodiment of this application is shown below. Figure 8 As shown, the device 800 includes: an acquisition module 810 and a display module 820; wherein, the acquisition module 810 is used to acquire display operations for comment information of a target song; the display module 820 is used to display the comment information of the target song in a first display area; and display the background information of the target song in a second display area; wherein, the first display area and the second display area are different display areas; the background information of the target song is generated based on at least one article related to the target song.

[0289] In some possible implementations, the background information of the target song is generated as follows: based on the tags of at least one real-time published article, music-related real-time published articles are initially selected from at least one real-time published article; based on the song title and artist name of the target song, historical published articles related to the target song are initially selected from at least one historical published article; and based on the basic information of the target song, the initially selected music-related real-time published articles, the initially selected historical published articles related to the target song, and the large language model, the background information of the target song is obtained.

[0290] In some feasible methods, the initial selection of music-related live articles is obtained by: selecting music-related live articles from at least one live article based on at least one of the word count and author information of each live article, as well as the tags of each live article.

[0291] In some feasible methods, the initial selection of historical publications related to the target song is obtained as follows: if at least one historical publication contains the target historical publication and the artist's name, then the target historical publication is selected as the initial selection of historical publications related to the target song.

[0292] In some feasible implementations, the background information of the target song is obtained as follows: extracting keywords from a preliminary selection of music-related real-time published articles and a preliminary selection of music-related historical published articles; matching the song title and artist name of the target song with the keywords from the preliminary selection of music-related real-time published articles and the preliminary selection of music-related historical published articles to select articles with successful keyword matching; and obtaining the background information of the target song based on the basic information of the target song, the articles with successful keyword matching, and a large language model.

[0293] In some implementations, the device 800 further includes a sending module 830, wherein the acquisition module 810 is further configured to acquire feedback operations on the background information of the target song to generate feedback information on the background information of the target song; and the sending module 840 is configured to send the feedback information to the backend server so that the backend server adjusts the background information of the target song based on the feedback information.

[0294] In some possible implementations, feedback information includes: content errors, irrelevance, low quality, violation, or other reasons.

[0295] In some implementations, the device 800 further includes: a receiving module 840, wherein the acquiring module 810 is further configured to acquire a click operation targeting the "Understand Original Article" identifier to generate a request message for understanding the original article; the sending module 830 is further configured to send a request message to a backend server to enable the backend server to retrieve the original article; and the receiving module 840 and the display module 820 are further configured to receive and display the original article respectively.

[0296] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 8 The device 800 shown can perform Figure 3 The corresponding method embodiments, and the foregoing and other operations and / or functions of each module in the device 800 are respectively for implementing Figure 2 For the sake of brevity, the corresponding processes in each method in the corresponding method embodiment will not be described in detail here.

[0297] The apparatus 800 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.

[0298] Figure 9 This is a schematic block diagram of the electronic device 900 provided in an embodiment of this application. The electronic device 900 may be a backend server corresponding to a music app or music interface, or a terminal device with a music app or music interface installed, but is not limited thereto. Figure 9 As shown, the electronic device 900 may include:

[0299] The system includes a memory 910 and a processor 920. The memory 910 stores a computer program 930 and transfers the computer program 930 to the processor 920. In other words, the processor 920 can retrieve and run the computer program 930 from the memory 910 to implement the methods described in the embodiments of this application.

[0300] For example, the processor 920 can be used to execute the steps in the above method according to the instructions in the computer program 930.

[0301] In some embodiments of this application, the processor 920 may include, but is not limited to:

[0302] General-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0303] In some embodiments of this application, the memory 910 includes, but is not limited to:

[0304] Volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0305] In some embodiments of this application, the computer program 930 may be divided into one or more modules, which are stored in the memory 910 and executed by the processor 920 to perform the method provided in this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 930 in the electronic device.

[0306] like Figure 9 As shown, the electronic device 900 may further include:

[0307] Transceiver 940, which can be connected to processor 920 or memory 910.

[0308] The processor 920 can control the transceiver 940 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 940 may include a transmitter and a receiver. The transceiver 940 may further include antennas, and the number of antennas may be one or more.

[0309] It should be understood that the various components in the electronic device 900 are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus.

[0310] According to one aspect of this application, a computer storage medium is provided that stores a computer program thereon, which, when executed by a computer, enables the computer to perform the methods of the above-described method embodiments. Alternatively, embodiments of this application also provide a computer program product containing instructions that, when executed by a computer, cause the computer to perform the methods of the above-described method embodiments.

[0311] According to another aspect of this application, a computer program product or computer program is provided, comprising 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 method described in the above-described method embodiments.

[0312] In other words, when implemented using software, it can be implemented wholly or partially in the form of a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. 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. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. 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 integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0313] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0314] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0315] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs. For example, the functional modules in the various embodiments of this application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0316] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for processing background information of a song, characterized in that, include: Obtain basic information about the target song; wherein, the basic information about the target song includes: the song title and the artist's name; Based on the tags of at least one real-time published article, music-related real-time published articles are initially selected from the at least one real-time published article; Based on the song title and artist name of the target song, select at least one historically published article that is related to the target song. Based on the basic information of the target song, the preliminary selection of real-time published articles related to music, the preliminary selection of historical published articles related to the target song, and the large language model, the background information of the target song is obtained.

2. The method according to claim 1, characterized in that, The preliminary selection of music-related real-time published articles from at least one real-time published article based on their respective tags includes: Based on at least one of the word count and author information of each of the at least one real-time published articles, and the tags of each of the at least one real-time published articles, music-related real-time published articles are initially selected from the at least one real-time published articles.

3. The method according to claim 2, characterized in that, The preliminary selection of music-related real-time published articles from the at least one real-time published article based on at least one of the word count and author information of each of the at least one real-time published article, and the tags of each of the at least one real-time published article, includes: If the tag of the first real-time published article among the at least one real-time published articles is related to music, and the word count of the first real-time published article is greater than a first value, then the first real-time published article is selected as the initial selection of music-related real-time published articles.

4. The method according to claim 2, characterized in that, The preliminary selection of music-related real-time published articles from the at least one real-time published article based on at least one of the word count and author information of each of the at least one real-time published article, and the tags of each of the at least one real-time published article, includes: If the tag of the second real-time published article among the at least one real-time published articles is related to music, and the author information of the second real-time published article is not in the blacklist, then the second real-time published article is selected as the initial selection of music-related real-time published articles.

5. The method according to claim 2, characterized in that, The preliminary selection of music-related real-time published articles from the at least one real-time published article based on at least one of the word count and author information of each of the at least one real-time published article, and the tags of each of the at least one real-time published article, includes: If the tag of the third real-time published article among the at least one real-time published articles is related to music, and the author information of the third real-time published article is not on the blacklist, and the word count of the third real-time published article is greater than the second value, then the third real-time published article is selected as the initial selection of music-related real-time published articles.

6. The method according to claim 1, characterized in that, The preliminary selection of historically published articles related to the target song from at least one historically published article, based on the song title and artist name of the target song, includes: If the target historical publication article in the at least one historical publication article includes the song title and artist name of the target song, then the target historical publication article is selected as the initial selection of historical publication articles related to the target song.

7. The method according to any one of claims 1-6, characterized in that, The background information of the target song is obtained based on its basic information, a preliminary selection of music-related real-time published articles, a preliminary selection of music-related historical published articles, and a large language model. This background information includes: Extract keywords from both a preliminary selection of real-time music-related articles and a preliminary selection of historical articles related to the target song; The song title and artist name of the target song are matched with keywords from both a preliminary selection of real-time music-related articles and a preliminary selection of historical music-related articles, and the articles with successful keyword matches are selected. Based on the basic information of the target song, the articles with successfully matched keywords, and the large language model, the background information of the target song is obtained.

8. The method according to claim 7, characterized in that, The background information of the target song is obtained based on the basic information of the target song, articles with successfully matched keywords, and the large language model, including: The basic information of the target song and the articles that successfully match the keywords are input into the large language model to obtain the background information of the target song.

9. The method according to claim 7, characterized in that, The background information of the target song is obtained based on the basic information of the target song, articles with successfully matched keywords, and the large language model, including: The basic information of the target song, the articles that successfully matched the keywords, and the prompt information are input into the large language model to obtain the background information of the target song.

10. The method according to claim 7, characterized in that, The background information of the target song is obtained based on the basic information of the target song, articles with successfully matched keywords, and the large language model, including: The content of articles that successfully match the song title and artist name of the target song with the keywords is matched to select paragraphs with successfully matched content. Based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model, the background information of the target song is obtained.

11. The method according to claim 10, characterized in that, The background information of the target song is obtained based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model, including: The basic information of the target song and the paragraphs that successfully match the content are input into the large language model to obtain the background information of the target song.

12. The method according to claim 10, characterized in that, The background information of the target song is obtained based on the basic information of the target song, the paragraphs with successfully matched content, and the large language model, including: The basic information of the target song, the paragraphs whose content was successfully matched, and the prompt information are input into the large language model to obtain the background information of the target song.

13. The method according to claim 11 or 12, characterized in that, Also includes: The paragraphs that successfully matched the content were cleaned.

14. The method according to any one of claims 1-6, characterized in that, The background information of the target song is obtained based on its basic information, a preliminary selection of music-related real-time published articles, a preliminary selection of music-related historical published articles, and a large language model. This background information includes: Based on the large language model, features are extracted from the initially selected real-time published articles related to music and the initially selected historical published articles related to the target song; Based on the large language model, the song title, artist name, and extracted article features of the target song are matched to select articles with successful feature matching. The articles that successfully match the features are processed based on the large language model to obtain the background information of the target song.

15. The method according to claim 14, characterized in that, The process of processing articles that have successfully matched features based on the large language model to obtain the background information of the target song includes: The background information of the target song is obtained by processing the articles and prompts that successfully match the features based on the large language model.

16. The method according to claim 9, 12, or 15, characterized in that, The prompt information includes at least one of the following: the task instruction to extract, the definition of the background information, the prompt information of the background information, the format requirements of the background information, positive examples, and negative examples.

17. A method for processing background information of a song, characterized in that, include: To retrieve and display comment information for the target song; The comment information for the target song is displayed in the first display area; The background information of the target song is displayed in the second display area; The first display area and the second display area are different display areas; The background information of the target song is generated based on at least one article related to the target song.

18. The method according to claim 17, characterized in that, The background information of the target song is generated in the following way: based on the tags of at least one real-time published article, real-time published articles related to music are initially selected from the at least one real-time published article; Based on the song title and artist name of the target song, a preliminary selection of historically published articles related to the target song is made from at least one historically published article. Based on the basic information of the target song, the preliminary selection of music-related real-time published articles, the preliminary selection of historically published articles related to the target song, and the large language model, the background information of the target song is obtained.

19. The method according to claim 17 or 18, characterized in that, Also includes: Obtain feedback on the background information of the target song to generate feedback information on the background information of the target song; The feedback information is sent to the backend server so that the backend server can adjust the background information of the target song based on the feedback information.

20. The method according to claim 17 or 18, characterized in that, Also includes: Get the click action for the "Learn More" tag to generate a request message for the "Learn More" article; Send the request message to the backend server so that the backend server can retrieve the original article; Receive and display the original article.

21. A device for processing background information of songs, characterized in that, include: Get modules, select modules, and generate modules; The acquisition module is used to acquire basic information about the target song; wherein, the basic information about the target song includes: the song title and the artist's name; The selection module is used to initially select music-related real-time published articles from the at least one real-time published article based on the tags of each of the at least one real-time published article; The selection module is also used to initially select historically published articles related to the target song from at least one historically published article based on the song name and artist name of the target song; The generation module is used to obtain the background information of the target song based on the basic information of the target song, the preliminarily selected real-time published articles related to music, the preliminarily selected historical published articles related to the target song, and the large language model.

22. A device for processing background information of songs, characterized in that, include: The acquisition module is used to retrieve and display comment information for the target song. The display module is used for: The comment information for the target song is displayed in the first display area; The background information of the target song is displayed in the second display area; The first display area and the second display area are different display areas; The background information of the target song is generated based on at least one article related to the target song.

23. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store a computer program, the processor being used to invoke and run the computer program stored in the memory to perform the method of any one of claims 1 to 20.

24. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the method as described in any one of claims 1 to 20.

25. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method as described in any one of claims 1 to 20.

Images