Musician recommendation method and device, electronic equipment and computer readable storage medium

Abstract

A musician recommendation method and device, electronic equipment and computer readable storage medium, relate to the technical field of computers. The method comprises: determining a first target song from a candidate song library based on a reference boutique song library, and determining a first target musician according to the musician information corresponding to the first target song; determining the song quality corresponding to the first target musician, and selecting a second target musician from the first target musician according to the song quality corresponding to the first target musician; determining the song recommendation index parameter corresponding to the second target musician, and performing musician recommendation according to the song recommendation index parameter corresponding to the second target musician. The present disclosure can realize the mining of boutique musicians by referring to the boutique song library for musician preliminary screening and the song quality for musician fine screening, and improve the possibility of users appreciating high-quality songs in various ways.

Description

Technical Field

[0001] The embodiments of this application relate to the field of computer technology, and more specifically, to a musician recommendation method, a musician recommendation device, an electronic device, and a computer-readable storage medium. Background Technology

[0002] This section is intended to provide background or context for embodiments of the present invention. The description herein is not intended to imply that it is prior art simply because it is included in this section.

[0003] Many music apps typically use music recommendation systems to provide users with music content to enjoy, thus satisfying their diverse music needs.

[0004] In related technologies, popular songs are usually statistically analyzed and distributed for recommendation. This method can create a strong head effect, which may result in many high-quality but not popular songs not reaching a wide range of users, narrowing the range of songs that users can appreciate and reducing the possibility of users enjoying high-quality songs in a diverse way.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] This application provides a musician recommendation method, a musician recommendation device, a computer-readable storage medium, and an electronic device, thereby at least partially solving the problem in related technologies that many high-quality but not popular songs cannot reach a wide range of users.

[0007] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0008] According to a first aspect of this application, a method for recommending musicians is provided. The method includes: determining a first target song from a candidate music library based on a reference library of high-quality music; determining a first target musician based on the musician information corresponding to the first target song; determining the quality of the song corresponding to the first target musician; selecting a second target musician from the first target musician based on the quality of the song corresponding to the first target musician; determining song recommendation index parameters corresponding to the second target musician; and recommending the musician based on the song recommendation index parameters corresponding to the second target musician.

[0009] In one exemplary embodiment of this application, determining the first target song from the candidate music library based on the reference high-quality music library includes: determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library; if the similarity between the first song to be processed and the target reference song meets a preset similarity threshold, then the first song to be processed is taken as the first target song.

[0010] In one exemplary embodiment of this application, determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library includes: extracting the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library based on a representation extraction model; comparing the representation features corresponding to the first song to be processed with the representation features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library; and determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library based on the similarity between the first song to be processed and each song in the reference high-quality music library.

[0011] In one exemplary embodiment of this application, before determining the first target song from the candidate music library based on the reference high-quality music library, the method further includes: determining abnormal songs in the reference high-quality music library and clearing the abnormal songs in the reference high-quality music library.

[0012] In one exemplary embodiment of this application, determining abnormal songs in the reference high-quality music library includes: extracting representational features corresponding to each song in the reference high-quality music library based on a representational extraction model; clustering the representational features corresponding to each song in the reference high-quality music library, and identifying songs with representational features outside the cluster as abnormal songs.

[0013] In one exemplary embodiment of this application, the representation extraction model is constructed in the following manner: a style tag model for style tag classification is trained based on song tag sample data to obtain a trained style tag model, wherein the song tag sample data includes style tag data corresponding to various types of songs; an encoding neural network branch for outputting a specified dimension is added to the trained style tag model to obtain a Siamese network model; the Siamese network model is trained based on song fragment sample data to obtain the representation extraction model.

[0014] In one exemplary embodiment of this application, the song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs, wherein the song fragments in the positively correlated sample data pairs belong to the same song, and the song fragments in the negatively correlated sample data pairs do not belong to the same song. Training the Siamese network model based on the song fragment sample data to obtain the representation extraction model includes: training the Siamese network model using a multi-task learning approach that includes label classification learning and metric learning based on the song fragment sample data to obtain the representation extraction model.

[0015] In one exemplary embodiment of this application, determining the song quality corresponding to the first target musician includes: determining the melody quality score and the audio quality score corresponding to the first target musician; and determining the song quality corresponding to the first target musician based on the melody quality score and the audio quality score.

[0016] In one exemplary embodiment of this application, determining the melody quality score and audio quality score corresponding to the first target musician includes: determining the melody quality score of the first target musician based on the melody information corresponding to one or more songs of the first target musician; and determining the audio quality score of the first target musician based on the audio information corresponding to one or more songs of the first target musician.

[0017] In one exemplary embodiment of this application, the step of selecting a second target musician from the first target musicians based on the song quality corresponding to the first target musician includes: if the song quality corresponding to the first target musician reaches a preset quality threshold, then the first target musician is selected as the second target musician.

[0018] In one exemplary embodiment of this application, the song recommendation index parameters include style tags and / or popularity scores. Determining the song recommendation index parameters corresponding to the second target musician includes: determining the style tags corresponding to one or more songs of the second target musician, and determining the style tag corresponding to the second target musician based on the style tags corresponding to one or more songs of the second target musician; and / or determining the popularity scores corresponding to one or more songs of the second target musician, and determining the popularity score corresponding to the second target musician based on the popularity scores corresponding to one or more songs of the second target musician.

[0019] In one exemplary embodiment of this application, determining the style tag corresponding to one or more songs of the second target musician includes: determining the style tag corresponding to the second song to be processed based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference music library, wherein the second song to be processed is a song of the second target musician.

[0020] In one exemplary embodiment of this application, determining the popularity score corresponding to one or more songs of the second target musician includes: determining the popularity score corresponding to the second song to be processed based on the popularity information corresponding to the second song to be processed, wherein the second song to be processed is a song of the second target musician.

[0021] In one exemplary embodiment of this application, the step of recommending musicians based on the song recommendation index parameters corresponding to the second target musician includes: determining musicians that match the user tags from the second target musicians based on the style tags corresponding to the second target musicians; and recommending musicians to the user objects corresponding to the user tags in descending order of popularity ratings corresponding to the musicians that match the user tags.

[0022] According to a second aspect of this application, a musician recommendation device is disclosed, the device comprising: a first screening module, configured to determine a first target song from a candidate music library based on a reference high-quality music library, and to determine a first target musician based on the musician information corresponding to the first target song; a second screening module, configured to determine the song quality corresponding to the first target musician, and to screen a second target musician from the first target musician based on the song quality corresponding to the first target musician; and a recommendation module, configured to determine the song recommendation index parameters corresponding to the second target musician, and to recommend the musician based on the song recommendation index parameters corresponding to the second target musician.

[0023] According to a third aspect of the embodiments of this application, an electronic device is disclosed, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the musician recommendation method and its possible implementations as disclosed in the first aspect.

[0024] According to a fourth aspect of the embodiments of this application, a computer program medium is disclosed, on which a computer program is stored, wherein when the computer program is executed by a processor, it implements the musician recommendation method and its possible implementations disclosed in the first aspect.

[0025] The technical solution of this application has the following beneficial effects:

[0026] In the aforementioned musician recommendation process, based on a reference to a high-quality music library, a first target song is determined from the candidate music library, and a first target musician is determined based on the musician information corresponding to the first target song; the quality of the songs corresponding to the first target musicians is determined, and a second target musician is selected from the first target musicians based on the song quality; the recommendation index parameters for the songs corresponding to the second target musicians are determined, and musician recommendations are made based on the song recommendation index parameters for the songs corresponding to the second target musicians. This disclosure, by conducting initial screening of musicians based on a reference to a high-quality music library and fine-tuning musicians based on song quality, can achieve the discovery and recommendation of high-quality musicians. This not only avoids, to some extent, the situation where many high-quality but not popular songs cannot reach a wide range of users due to the head effect, thus expanding the range of songs that users can appreciate and increasing the possibility of users appreciating high-quality songs in a diverse way, but also discovers high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0027] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0029] Figure 1 This illustration shows a flowchart of a musician recommendation method implemented in this exemplary embodiment;

[0030] Figure 2 This illustration shows a flowchart of a process for determining a first target song from a candidate music library in this exemplary embodiment;

[0031] Figure 3 This exemplary embodiment shows a schematic diagram of the song distribution in a reference music library;

[0032] Figure 4A This exemplary embodiment shows a schematic diagram of the song distribution in the reference music library before abnormal song removal;

[0033] Figure 4B This exemplary embodiment shows a schematic diagram of the song distribution in the reference music library after abnormal songs have been cleared.

[0034] Figure 5This illustration shows a flowchart of a representation extraction model in this exemplary embodiment;

[0035] Figure 6 This illustration shows a flowchart of a process for determining the quality of a song corresponding to a first target musician in this exemplary embodiment.

[0036] Figure 7 This illustration shows a flowchart of a process for discovering and recommending outstanding musicians in this exemplary embodiment;

[0037] Figure 8 This diagram illustrates a structural block diagram of a musician recommendation device according to this exemplary embodiment;

[0038] Figure 9 An electronic device for implementing the musician recommendation method described above is shown in this exemplary embodiment.

[0039] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation

[0040] The principles and spirit of this application will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement this application, and are not intended to limit the scope of this application in any way. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0041] Those skilled in the art will understand that the embodiments of this application can be implemented as an apparatus, device, method, or computer program product. Therefore, this application can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0042] According to embodiments of this application, a musician recommendation method, a musician recommendation device, an electronic device, and a computer-readable storage medium are proposed.

[0043] The number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0044] The principles and spirit of this application will be explained in detail below with reference to several representative embodiments. Invention Overview

[0046] In the related technologies of this application, the distribution and recommendation of top popular songs may result in many high-quality but not popular songs not reaching a wide range of users, narrowing the range of songs that users can appreciate and reducing the possibility of users appreciating high-quality songs in a diverse way.

[0047] Based on the above problems, this application proposes a musician recommendation method. By referring to a high-quality music library for initial screening of musicians and then further screening them based on song quality, this method can not only avoid the problem of many high-quality but not popular songs failing to reach a wide range of users due to the head effect, thus expanding the range of songs that users can enjoy and increasing the possibility of users appreciating high-quality songs in a diverse way, but also discover high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0048] Application Scenarios Overview

[0049] It should be noted that the following application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the implementation of this application is not limited in any way. On the contrary, the implementation of this application can be applied to any applicable scenario. For example, it can be applied to the song recommendation scenario of music software.

[0050] Exemplary methods

[0051] The following section, in conjunction with the application scenarios described above, provides reference... Figure 1 This application describes a musician recommendation method based on an exemplary embodiment.

[0052] Please see Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a musician recommendation method according to an example embodiment of this application. Figure 1 As shown, the musician's recommendation method may include:

[0053] Step S110: Based on the reference high-quality music library, determine the first target song from the candidate music library, and determine the first target musician according to the musician information corresponding to the first target song;

[0054] Step S120: Determine the quality of the songs corresponding to the first target musician, and select the second target musician from the first target musicians based on the quality of the songs corresponding to the first target musicians;

[0055] Step S130: Determine the song recommendation index parameters corresponding to the second target musician, and recommend musicians based on the song recommendation index parameters corresponding to the second target musician.

[0056] Implementation Figure 1The musician recommendation method shown here uses a high-quality music library for initial screening of musicians and then further refines the screening based on song quality. This method can discover and recommend high-quality musicians, which can not only avoid the problem of many high-quality but not popular songs not reaching a wide range of users due to the head effect, thus expanding the range of songs that users can enjoy and increasing the possibility of users appreciating high-quality songs in a diverse way, but also discover high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0057] The "head effect" here refers to the fact that songs with high play counts and popularity, and high rankings, typically attract a large audience. Long-tail songs, on the other hand, refer to songs with low play counts and popularity, and lower rankings; these songs are usually difficult for listeners to discover.

[0058] These steps are described in detail below.

[0059] In step S110, based on the reference high-quality music library, the first target song is determined from the candidate music library, and the first target musician is determined according to the musician information corresponding to the first target song.

[0060] The "reference library of high-quality songs" refers to a library of high-quality songs of various types, which can be pre-built. For example, the reference library of high-quality songs may include upbeat songs, sad songs, piano solo songs, etc., and this disclosure does not make specific limitations in this regard.

[0061] Optionally, the reference premium music library can be constructed from multiple sub-libraries, each containing different types of premium songs. Specifically, mood type, scene type, theme type, etc., can be used as the criteria for distinguishing between different sub-libraries. For example, the reference premium music library can be constructed from multiple sub-libraries such as upbeat sub-libraries, sad sub-libraries, and piano solo sub-libraries; this disclosure does not impose specific limitations on this. Songs in the reference premium music library can have the same preset style tags as the type of their respective sub-libraries. Taking an upbeat sub-library as an example, the songs in this sub-library can have preset upbeat style tags.

[0062] By constructing a reference library of high-quality songs using different types of sub-libraries, the types of songs included in the reference library can be enriched, and songs with the characteristics of high-quality songs can be selected from the candidate library more comprehensively and accurately.

[0063] The candidate music library refers to a collection of songs published by musicians that can be recommended to users, and may include songs referenced from the premium music library. Understandably, in practical applications, due to differences in musicians' styles and quality, the candidate music library may contain songs of different types and qualities.

[0064] The first target song refers to a song selected from the candidate song library that possesses the characteristics of songs in the high-quality song library.

[0065] In one alternative implementation, the first target song is determined from the candidate music library based on a reference library of high-quality music, such as... Figure 2 As shown, this can be achieved through the following steps:

[0066] Step S210: Determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library;

[0067] Step S220: If the similarity between the first song to be processed and the target reference song meets the preset similarity threshold, then the first song to be processed is taken as the first target song.

[0068] Figure 2 In the steps shown, by selecting the reference song with the highest similarity to the song in the candidate song library from the reference boutique library and comparing the similarity, it can be determined to some extent whether the songs in the candidate song library have the characteristics of boutique songs.

[0069] Specifically, in step S210, a target reference song with the highest similarity to the first song to be processed is determined from the reference music library, wherein the first song to be processed is any song in the candidate music library.

[0070] The target reference song can be the song with the highest similarity to the first song to be processed, selected from the reference library of high-quality songs.

[0071] Optionally, the above-mentioned determination of the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library can be achieved through the following steps: based on the representation extraction model, extract the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library; compare the representation features corresponding to the first song to be processed with the representation features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library; based on the similarity between the first song to be processed and each song in the reference high-quality music library, determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library.

[0072] Among them, the representation extraction model can be used to extract the representational features of songs.

[0073] Specifically, key audio information of each song in the reference high-quality music library can be extracted in advance to obtain key audio information corresponding to each song in the reference high-quality music library. Then, a representation extraction model is used to extract representation features of key audio information corresponding to each song in the reference high-quality music library to obtain representation features of each song in the reference high-quality music library, thereby improving the efficiency of feature extraction.

[0074] Correspondingly, the key audio information of the first song to be processed can be extracted in advance to obtain the key audio information of the first song to be processed. Then, the representation extraction model is used to extract the representation features of the key audio information corresponding to the first song to be processed to obtain the representation features of the first song to be processed.

[0075] The key audio information may include the audio information of the verse and / or chorus of the song, which can be obtained through the following steps: Segment the song to identify the verse, chorus, and interlude, and determine the timing information of each part, thereby extracting the audio information of the verse and / or chorus. For example, a deep learning algorithm can be used to learn the start times of the verse and chorus from the labeled verse and chorus timing data, and then extract the audio information of the verse and chorus. Alternatively, a clustering algorithm can be used to find the most frequently repeated frames or song segments, and based on this, determine the chorus and extract its audio information.

[0076] After obtaining the representational features corresponding to the first song to be processed and the representational features corresponding to each song in the reference high-quality music library, the representational features corresponding to the first song to be processed can be compared with the representational features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library. The song with the highest similarity to the first song to be processed in the reference high-quality music library is taken as the target reference song so as to measure the similarity between the songs based on the similarity parameter and provide a reference object for the identification of high-quality songs.

[0077] Specifically, in step S220, if the similarity between the first song to be processed and the target reference song meets the preset similarity threshold, then the first song to be processed is taken as the first target song.

[0078] The preset similarity threshold can be set based on experience or needs, and this disclosure does not impose specific limitations on it. If the similarity between the first song to be processed and the target reference song meets the preset similarity threshold, it indicates that the first song to be processed and the target reference song have relatively similar characteristics. If the target reference song is a song in the reference high-quality music library, the first song to be processed has the characteristics of a high-quality song, and at this time, the first song to be processed can be used as the first target song.

[0079] In one alternative implementation, before determining the first target song from the candidate music library based on the reference high-quality music library, the following steps may be performed: identifying abnormal songs in the reference high-quality music library and removing abnormal songs from the reference high-quality music library.

[0080] Abnormal songs refer to songs in the reference library whose audio data is abnormal. By removing abnormal songs from the reference library in advance, the quality of the reference library can be improved to a certain extent, thereby ensuring the accuracy of the first target song determined based on the reference library.

[0081] In one optional implementation, the above-mentioned determination of abnormal songs in the reference high-quality music library can be achieved through the following steps: extracting the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model; clustering the representation features corresponding to each song in the reference high-quality music library, and identifying songs with out-of-class representation features as abnormal songs.

[0082] Specifically, the representative features of each song in the reference high-quality music library can be clustered to determine the cluster center. Songs with representative features that are far from the cluster center are identified as abnormal songs, thereby realizing the identification of abnormal songs and further improving the quality of songs in the reference high-quality music library.

[0083] For example, such as Figure 3 As shown, a schematic diagram of song distribution based on a high-quality music library is provided. In this diagram, the same number represents songs in the same sub-library. Songs with characteristics that are far from the cluster center can be regarded as abnormal songs.

[0084] If the referenced premium music library contains multiple sub-libraries, for example, Figure 4A A schematic diagram showing the distribution of songs in the premium music library before abnormal songs are removed is provided. Figure 4B A schematic diagram of the song distribution in the premium music library after removing abnormal songs is provided, making the boundaries of song distribution in different sub-libraries clearer.

[0085] In one alternative implementation, such as Figure 5 As shown, the representation extraction model is constructed in the following manner:

[0086] Step S510: Train a style tag model for style tag classification based on song tag sample data to obtain the trained style tag model. The song tag sample data contains style tag data corresponding to various types of songs.

[0087] Step S520: Add a branch of the encoding neural network for outputting a specified dimension to the trained style label model to obtain the Siamese network model.

[0088] Step S530: Train the Siamese network model based on song fragment sample data to obtain the representation extraction model.

[0089] Figure 5 In the steps shown, a Siamese network model is constructed, and the trained Siamese network model is used for feature extraction. The model has strong stability and generalization ability, which can ensure the accuracy of feature extraction to a certain extent.

[0090] Specifically, in step S510, a style tag model for style tag classification is trained based on the song tag sample data to obtain the trained style tag model. The song tag sample data contains style tag data corresponding to various types of songs.

[0091] The style tag model can include a basic neural network and a classification neural network. The basic neural network can convert the input sample data into a vector of a set dimension, and the classification neural network can output classification information corresponding to at least one tag category based on the vector input from the basic neural network.

[0092] Specifically, in step S520, an encoding neural network branch for outputting a specified dimension is added to the trained style tag model to obtain a Siamese network model.

[0093] The encoding neural network can output a vector of a specified dimension based on the vector input from the base neural network. Optionally, the encoding neural network can be implemented using a fully connected network.

[0094] Specifically, in step S530, the Siamese network model is trained based on song fragment sample data to obtain a representation extraction model.

[0095] Optionally, the song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs. Positively correlated sample data pairs contain song fragments belonging to the same song, while negatively correlated sample data pairs contain song fragments not belonging to the same song. The Siamese network model is trained based on the song fragment sample data to obtain the representation extraction model, which can be achieved through the following steps: Based on the song fragment sample data, the Siamese network model is trained using a multi-task learning approach that includes label classification learning and metric learning to obtain the representation extraction model.

[0096] Optionally, the loss function of the characterization extraction model can be optimized using the cross-entropy loss function and the contrastive learning loss function. The cross-entropy loss function is a loss function in deep learning that measures the difference between the predicted result distribution and the true labeled distribution; contrastive learning is a type of self-supervised learning that does not rely on labeled data. An example of a contrastive learning loss function is the NT-Xent loss function, as shown in equation (1).

[0097] (1)

[0098] in, For loss function, and This represents two random segments of the same song, and is a positively correlated example. and z k For two random segments from different songs, is a negatively correlated example; parameters Used to control the sensitivity of the loss function to negative sample pairs, the subscripts i, j, k are the identifiers of the corresponding song segments, and N is the number of samples in the batch.

[0099] Specifically, during use, the representation extraction model can ultimately output a vector of a specific dimension as the representation feature of the corresponding song.

[0100] Furthermore, after determining the primary target song, the primary target musician can be identified based on the musician information corresponding to the primary target song. This musician information can be the song's creator or publisher; the primary target musician refers to the musician associated with the primary target song.

[0101] After determining the first target musician using step S110, step S120 can be executed.

[0102] In step S120, the quality of the song corresponding to the first target musician is determined, and a second target musician is selected from the first target musicians based on the quality of the song corresponding to the first target musician.

[0103] Song quality may include, but is not limited to, melody quality and audio quality.

[0104] In one alternative implementation, the above-mentioned determination of the song quality corresponding to the first target musician, such as... Figure 6 As shown, the following steps may be included:

[0105] Step S610: Determine the melody quality score and audio quality score corresponding to the first target musician;

[0106] Step S620: Determine the song quality corresponding to the first target musician based on the melody quality score and audio quality score.

[0107] Figure 6 The steps shown demonstrate a comprehensive and objective approach to determining the quality of a target musician's song by considering both melody quality and audio quality.

[0108] Specifically, in step S610, the melody quality score and audio quality score corresponding to the first target musician are determined.

[0109] The melody quality score and audio quality score are the quality scores for the song's melody and audio, respectively. For example, the more complex the melody, the higher the melody quality score; the higher the audio integrity and the less noise, the higher the audio quality score. It should be noted that in practical applications, melody quality and audio quality can also be scored based on other parameters, but this disclosure does not specifically limit this.

[0110] In one optional implementation, the determination of the melody quality score and audio quality score corresponding to the first target musician can be achieved through the following steps: determining the melody quality score of the first target musician based on the melody information corresponding to one or more songs of the first target musician; determining the audio quality score of the first target musician based on the audio information corresponding to one or more songs of the first target musician.

[0111] For example, the melody information of the top P songs with the highest play count of the first target musician can be obtained, and the melody quality score of the first target musician can be determined based on the melody information of these P songs, where P is the number of songs selected from the songs of the first target musician for melody quality scoring, which can be set according to the actual situation, and this disclosure does not make specific limitations in this regard.

[0112] Specifically, the melody quality scores of these P songs can be determined individually, and based on these scores, the melody quality score of the first target musician can be determined. For example, the melody quality scores of the P songs can be averaged or summed to obtain the melody quality score of the first target musician. By comprehensively scoring the melody quality of multiple songs, the accuracy of the melody quality score for the first target musician can be ensured to a certain extent.

[0113] For example, the audio information of the top Q songs with the highest play count of the first target musician can be obtained, and the audio quality score of the first target musician can be determined based on the audio information of these Q songs, where Q is the number of songs selected from the songs of the first target musician for audio quality scoring, which can be set according to the actual situation, and this disclosure does not make specific limitations in this regard.

[0114] Specifically, the audio quality scores of the Q songs can be determined individually, and based on these scores, the audio quality score of the first target musician can be determined. For example, the audio quality scores of the Q songs can be averaged or summed to obtain the audio quality score of the first target musician. By comprehensively scoring the audio quality of multiple songs, the accuracy of the audio quality score for the first target musician can be ensured to a certain extent.

[0115] Specifically, in step S620, the song quality corresponding to the first target musician is determined based on the melody quality score and the audio quality score.

[0116] For example, the quality of the song corresponding to the first target musician can be evaluated by averaging or summing the melody quality score and audio quality score of the first target musician and obtaining the final score.

[0117] After determining the quality of the songs of the first target musician, the first target musician can be further screened to obtain the second target musician.

[0118] In one optional implementation, the above-mentioned selection of a second target musician from the first target musicians based on the song quality corresponding to the first target musician can be achieved through the following steps: if the song quality corresponding to the first target musician reaches a preset quality threshold, then the first target musician is selected as the second target musician.

[0119] The preset quality threshold can be set in advance based on experience, and no specific limitations are made here. A second target musician is selected from the first target musicians whose song quality exceeds the preset quality threshold, in order to discover high-quality musicians.

[0120] After identifying the second target musician, step S130 can be further executed.

[0121] In step S130, the song recommendation index parameters corresponding to the second target musician are determined, and musician recommendations are made based on the song recommendation index parameters corresponding to the second target musician.

[0122] The song recommendation metrics may include style tags and / or popularity ratings.

[0123] In one optional implementation, the above-mentioned determination of the song recommendation index parameters corresponding to the second target musician can be achieved through the following steps: determining the style tags corresponding to one or more songs of the second target musician, and determining the style tags corresponding to the second target musician based on the style tags corresponding to one or more songs of the second target musician; and / or determining the popularity scores corresponding to one or more songs of the second target musician, and determining the popularity scores corresponding to the second target musician based on the popularity scores corresponding to one or more songs of the second target musician.

[0124] The style tags can be tags for music genres, such as jazz, blues, rock, pop, etc.; they can also be tags for moods, such as sad, cheerful, excited, etc.; they can also be tags for scenes, such as gym, bar, coffee shop, etc.; and they can also be tags for themes, such as sleep aid, night, dream, etc. This disclosure does not impose specific limitations on these.

[0125] In one alternative implementation, determining the style tag corresponding to one or more songs by the second target musician can be achieved through the following steps: based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference boutique music library, determine the style tag corresponding to the second song to be processed, wherein the second song to be processed is a song by the second target musician.

[0126] Optionally, a preset number of songs can be selected from the songs of the second target musician as the second set of songs to be processed. Specifically, the second set of songs to be processed can be selected based on the number of plays of the songs of the second target musician.

[0127] Songs in the reference music library can be configured with corresponding preset style tags. The preset style tag of the target reference song with the highest similarity to the second song to be processed can be used as the style tag of the second song to be processed, thus obtaining the style tag of the second song to be processed. This is a relatively convenient and quick method.

[0128] After identifying the style tags corresponding to one or more songs by the second target musician, tag statistics can be performed according to tag type to determine the style tags corresponding to the second target musician. For example, the style tag with the most statistical occurrences can be used as the style tag corresponding to the second target musician.

[0129] In one alternative implementation, determining the popularity score corresponding to one or more songs of the second target musician can be achieved through the following steps: determining the popularity score corresponding to the second song to be processed based on the popularity information of the second song to be processed, wherein the second song to be processed is a song of the second target musician.

[0130] The popularity information can include data such as the number of likes and comments. Optionally, a preset number of songs can be selected from the songs of the second target musician as the second songs to be processed. Specifically, the second songs to be processed can be selected based on the number of plays of the songs of the second target musician. The popularity score of the second songs to be processed can be determined by obtaining the popularity information corresponding to them. For example, the more likes and comments, the higher the popularity score; the fewer likes and comments, the lower the popularity score, which is relatively convenient and quick to implement.

[0131] After determining the popularity ratings of one or more songs by the second target musician, the popularity rating of the second target musician can be obtained by statistically analyzing the popularity ratings of each song, such as by summing or averaging.

[0132] In one optional implementation, the musician recommendation based on the song recommendation index parameters corresponding to the second target musician can be achieved through the following steps: based on the style tags corresponding to the second target musician, determine the musicians that match the user tags from the second target musicians; and recommend musicians to the user objects corresponding to the user tags in descending order of popularity scores corresponding to the musicians that match the user tags.

[0133] For example, if a user has a fitness tag, a second musician with a fitness style tag can be recommended to that user in descending order of popularity rating. By taking into account style factors, personalized recommendations of high-quality musicians can be achieved.

[0134] Furthermore, a hit prediction model can be used to predict the hit probability of one or more songs by a second target musician. Based on the hit probability of one or more songs by the second target musician and the basic traffic data of the second target musician, the hit probability of the second target musician can be obtained. Second musicians who match user tags are recommended to users in order of hit probability from high to low.

[0135] like Figure 7 As shown, a flowchart for discovering and recommending outstanding musicians is provided, which may include the following steps:

[0136] Step S701: Based on the representation extraction model, extract the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference music library, wherein the first song to be processed is any song in the candidate music library.

[0137] Step S702: Compare the representation features corresponding to the first song to be processed with the representation features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library.

[0138] Step S703: Based on the similarity between the first song to be processed and each song in the reference high-quality music library, determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library;

[0139] Step S704: If the similarity between the first song to be processed and the target reference song meets the preset similarity threshold, then the first song to be processed is taken as the first target song.

[0140] Step S705: Determine the first target musician based on the musician information corresponding to the first target song;

[0141] Step S706: Determine the melody quality score and audio quality score corresponding to the first target musician, and determine the song quality corresponding to the first target musician based on the melody quality score and audio quality score;

[0142] Step S707: If the quality of the song corresponding to the first target musician reaches the preset quality threshold, then the first target musician is used as the second target musician.

[0143] Step S708: Determine the song recommendation index parameters corresponding to the second target musician, and recommend musicians based on the song recommendation index parameters corresponding to the second target musician.

[0144] The embodiments of this application, by initially screening musicians based on a high-quality music library and then further screening them based on song quality, can achieve the discovery and recommendation of high-quality musicians. This not only avoids, to some extent, the situation where many high-quality but not popular songs cannot reach a wide range of users due to the head effect, thus expanding the range of songs that users can appreciate and increasing the possibility of users appreciating high-quality songs in a diverse way, thereby improving the quality of music recommendations, but also discovers high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0145] Exemplary media

[0146] After introducing the methods of exemplary embodiments of this application, the media of exemplary embodiments of this application will now be described.

[0147] In some possible implementations, various aspects of this application may also be implemented as a medium storing a program product capable of implementing the musician recommendation method described above. In some possible implementations, various aspects of this application may also be implemented as a program product containing program code that, when run on an electronic device, causes the electronic device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0148] Specifically, when the processor of the electronic device executes the program code, it can be used to implement the following steps:

[0149] Based on the reference library of high-quality songs, the first target song is determined from the candidate song library, and the first target musician is determined based on the musician information corresponding to the first target song;

[0150] Determine the quality of the songs corresponding to the first target musician, and select the second target musician from the first target musicians based on the quality of the songs corresponding to the first target musicians;

[0151] Determine the song recommendation index parameters corresponding to the second target musician, and recommend musicians based on the song recommendation index parameters corresponding to the second target musician.

[0152] In an optional implementation, based on the above scheme, the determination of the first target song from the candidate music library based on the reference high-quality music library can be achieved through the following steps: determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library; if the similarity between the first song to be processed and the target reference song meets a preset similarity threshold, then the first song to be processed is taken as the first target song.

[0153] In an optional implementation, based on the above scheme, determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library can be achieved through the following steps: extracting the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model; comparing the representation features corresponding to the first song to be processed with the representation features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library; and determining the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library based on the similarity between the first song to be processed and each song in the reference high-quality music library.

[0154] In an alternative implementation, based on the above scheme, before determining the first target song from the candidate song library based on the reference high-quality song library, the following steps can be taken: identifying abnormal songs in the reference high-quality song library and clearing the abnormal songs in the reference high-quality song library.

[0155] In an optional implementation, based on the above scheme, the abnormal songs in the reference high-quality music library can be determined by the following steps: extracting the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model; clustering the representation features corresponding to each song in the reference high-quality music library, and identifying the songs with representation features outside the cluster as abnormal songs.

[0156] In one optional implementation, based on the above scheme, the representation extraction model is constructed in the following way: a style tag model for style tag classification is trained based on song tag sample data to obtain a trained style tag model, wherein the song tag sample data contains style tag data corresponding to various types of songs; an encoding neural network branch for outputting a specified dimension is added to the trained style tag model to obtain a Siamese network model; the Siamese network model is trained based on song fragment sample data to obtain the representation extraction model.

[0157] In an optional implementation, based on the above scheme, the song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs, wherein the song fragments in the positively correlated sample data pairs belong to the same song, and the song fragments in the negatively correlated sample data pairs do not belong to the same song. The Siamese network model is trained based on the song fragment sample data to obtain the representation extraction model, which can be achieved through the following steps: Based on the song fragment sample data, the Siamese network model is trained using a multi-task learning method that includes label classification learning and metric learning to obtain the representation extraction model.

[0158] In one alternative implementation, the quality of the song corresponding to the first target musician can be determined based on the above scheme by the following steps: determining the melody quality score and audio quality score corresponding to the first target musician; and determining the quality of the song corresponding to the first target musician based on the melody quality score and audio quality score.

[0159] In one optional implementation, based on the above scheme, determining the melody quality score and audio quality score corresponding to the first target musician can be achieved through the following steps: determining the melody quality score of the first target musician based on the melody information corresponding to one or more songs of the first target musician; determining the audio quality score of the first target musician based on the audio information corresponding to one or more songs of the first target musician.

[0160] In an optional implementation, based on the above scheme, the selection of a second target musician from the first target musicians according to the quality of the songs corresponding to the first target musicians can be achieved through the following steps: if the quality of the songs corresponding to the first target musicians reaches a preset quality threshold, then the first target musicians are selected as the second target musicians.

[0161] In an optional implementation, based on the above scheme, the song recommendation index parameters include style tags and / or popularity scores. Determining the song recommendation index parameters corresponding to the second target musician can be achieved through the following steps: determining the style tags corresponding to one or more songs of the second target musician, and determining the style tags corresponding to the second target musician based on the style tags corresponding to one or more songs of the second target musician; and / or determining the popularity scores corresponding to one or more songs of the second target musician, and determining the popularity scores corresponding to the second target musician based on the popularity scores corresponding to one or more songs of the second target musician.

[0162] In an optional implementation, based on the above scheme, determining the style tag corresponding to one or more songs of the second target musician can be achieved through the following steps: based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference boutique music library, the style tag corresponding to the second song to be processed is determined, wherein the second song to be processed is a song of the second target musician.

[0163] In one alternative implementation, based on the above scheme, determining the popularity score corresponding to one or more songs of the second target musician can be achieved through the following steps: determining the popularity score corresponding to the second song to be processed based on the popularity information of the second song to be processed, wherein the second song to be processed is a song of the second target musician.

[0164] In an optional implementation, based on the above scheme, musician recommendations are made according to the song recommendation index parameters corresponding to the second target musician. This can be achieved through the following steps: based on the style tags corresponding to the second target musician, musicians matching the user tags are determined from the second target musicians; musicians are recommended to the user objects corresponding to the user tags in descending order of popularity ratings corresponding to the musicians matching the user tags.

[0165] The embodiments of this application, by initially screening musicians based on a high-quality music library and then further screening them based on song quality, can achieve the discovery and recommendation of high-quality musicians. This not only avoids, to some extent, the situation where many high-quality but not popular songs cannot reach a wide range of users due to the head effect, thus expanding the range of songs that users can appreciate and increasing the possibility of users appreciating high-quality songs in a diverse way, thereby improving the quality of music recommendations, but also discovers high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0166] It should be noted that the aforementioned medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0167] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0168] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, RF (Radio Frequency), or any suitable combination thereof.

[0169] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0170] Exemplary device

[0171] After introducing the medium of exemplary embodiments of this application, the following references are made. Figure 8 The musician recommendation device according to an exemplary embodiment of this application will be described.

[0172] Please see Figure 8 , Figure 8 The diagram shown is a structural block diagram of a musician recommendation device according to an example embodiment of this application. Figure 8 As shown, a musician recommendation device 800 according to an example embodiment of this application includes: a first screening module 810, a second screening module 820, and a recommendation module 830, wherein:

[0173] The first screening module 810 is used to determine the first target song from the candidate music library based on the reference high-quality music library, and to determine the first target musician based on the musician information corresponding to the first target song;

[0174] The second filtering module 820 is used to determine the quality of the songs corresponding to the first target musicians, and to filter the second target musicians from the first target musicians based on the quality of the songs corresponding to the first target musicians.

[0175] The recommendation module 830 is used to determine the song recommendation index parameters corresponding to the second target musician, and to recommend musicians based on the song recommendation index parameters corresponding to the second target musician.

[0176] In an optional implementation, based on the aforementioned scheme, the first screening module 810 further includes: a target reference song determination module, used to determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library; and a first target song identification module, used to identify the first song to be processed as the first target song if the similarity between the first song to be processed and the target reference song meets a preset similarity threshold.

[0177] In an optional implementation, based on the aforementioned scheme, the target reference song determination module may be configured to: extract the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model; compare the representation features corresponding to the first song to be processed with the representation features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library; and determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library based on the similarity between the first song to be processed and each song in the reference high-quality music library.

[0178] In an alternative implementation, based on the aforementioned scheme, before determining the first target song from the candidate song library based on the reference high-quality song library, the musician recommendation device 800 may further include: an abnormal song removal module, used to identify abnormal songs in the reference high-quality song library and remove abnormal songs in the reference high-quality song library.

[0179] In an optional implementation, based on the aforementioned scheme, the abnormal song removal module further includes: a representation feature determination module, used to extract the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model; and an abnormal song identification module, used to cluster the representation features corresponding to each song in the reference high-quality music library, and identify songs with representation features outside the cluster as abnormal songs.

[0180] In an optional implementation, based on the aforementioned scheme, the musician recommendation device 800 may further include: a representation extraction model construction module. The representation extraction model construction module includes: a style tag model training module, used to train a style tag model for style tag classification based on song tag sample data, to obtain a trained style tag model; the song tag sample data includes style tag data corresponding to various types of songs; a Siamese network model construction module, used to add an encoding neural network branch for outputting a specified dimension to the trained style tag model, to obtain a Siamese network model; and a Siamese network model training module, used to train the Siamese network model based on song fragment sample data, to obtain a representation extraction model.

[0181] In an optional implementation, based on the aforementioned scheme, the song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs, wherein the song fragments in the positively correlated sample data pairs belong to the same song, and the song fragments in the negatively correlated sample data pairs do not belong to the same song. The Siamese network model training module can be configured to: train the Siamese network model based on the song fragment sample data using a multi-task learning method that includes label classification learning and metric learning, to obtain a representation extraction model.

[0182] In an optional implementation, based on the aforementioned scheme, the second screening module 820 includes: a quality scoring module, used to determine the melody quality score and audio quality score corresponding to the first target musician; and a quality determination module, used to determine the song quality corresponding to the first target musician based on the melody quality score and audio quality score.

[0183] In an optional implementation, based on the aforementioned scheme, the quality scoring module further includes: a melody quality scoring module, used to determine the melody quality score of the first target musician based on the melody information corresponding to one or more songs of the first target musician; and an audio quality scoring module, used to determine the audio quality score of the first target musician based on the audio information corresponding to one or more songs of the first target musician.

[0184] In an optional implementation, based on the aforementioned scheme, the second screening module 820 further includes: a high-quality musician determination module, used to determine the first target musician as the second target musician if the quality of the song corresponding to the first target musician reaches a preset quality threshold.

[0185] In an optional implementation, based on the aforementioned scheme, the song recommendation index parameters include style tags and / or popularity scores. The recommendation module 830 includes: a musician style determination module, used to determine the style tags corresponding to one or more songs of the second target musician, and to determine the style tags corresponding to the second target musician based on the style tags corresponding to one or more songs of the second target musician; and / or a musician popularity determination module, used to determine the popularity scores corresponding to one or more songs of the second target musician, and to determine the popularity scores corresponding to the second target musician based on the popularity scores corresponding to one or more songs of the second target musician.

[0186] In an optional implementation, based on the aforementioned scheme, the musician style determination module includes: a song style determination module, used to determine the style tag corresponding to the second song to be processed based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference high-quality music library, wherein the second song to be processed is a song by the second target musician.

[0187] In one optional implementation, based on the aforementioned scheme, the musician popularity determination module includes a song popularity determination module, used to determine the popularity score corresponding to the second song to be processed based on the popularity information corresponding to the second song to be processed, wherein the second song to be processed is a song of the second target musician.

[0188] In an optional implementation, based on the aforementioned scheme, the recommendation module 830 further includes: a musician style matching module, used to determine musicians that match the user tags from the second target musicians based on the style tags corresponding to the second target musicians; and a musician recommendation module, used to recommend musicians to the user objects corresponding to the user tags in descending order of popularity ratings corresponding to the musicians that match the user tags.

[0189] The embodiments of this application, by initially screening musicians based on a high-quality music library and then further screening them based on song quality, can achieve the discovery and recommendation of high-quality musicians. This not only avoids, to some extent, the situation where many high-quality but not popular songs cannot reach a wide range of users due to the head effect, thus expanding the range of songs that users can appreciate and increasing the possibility of users appreciating high-quality songs in a diverse way, thereby improving the quality of music recommendations, but also discovers high-quality musicians in long-tail songs, providing them with more exposure and increasing the diversity of recommended songs to a certain extent.

[0190] It should be noted that although several modules or units of the musician recommendation device have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0191] Exemplary electronic devices

[0192] Having described the musician recommendation method, medium, and apparatus according to exemplary embodiments of this application, we will now describe an electronic device according to another exemplary embodiment of this application.

[0193] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0194] The following reference Figure 9 To describe an electronic device 900 according to such an embodiment of the present invention. Figure 9 The electronic device 900 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0195] like Figure 9 As shown, the electronic device 900 is presented in the form of a general-purpose computing device. The components of the electronic device 900 may include, but are not limited to: at least one processing unit 910, at least one storage unit 920, a bus 930 connecting different system components (including storage unit 920 and processing unit 910), and a display unit 940.

[0196] The storage unit stores program code that can be executed by the processing unit 910, causing the processing unit 910 to perform the steps described in the "Exemplary Methods" section above, according to various exemplary embodiments of the present invention.

[0197] Storage unit 920 may include readable media in the form of volatile storage units, such as random access memory (RAM) 921 and / or cache memory 922, and may further include read-only memory (ROM) 923.

[0198] Storage unit 920 may also include a program / utility 924 having a set (at least one) of program modules 925, such program modules 925 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may contain the reality of the network environment.

[0199] Bus 930 may include a data bus, an address bus, and a control bus.

[0200] Electronic device 900 can also communicate with one or more external devices 990 (e.g., keyboard, pointing device, Bluetooth device, etc.) via input / output (I / O) interface 950. Furthermore, electronic device 900 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 960. Figure 9 As shown, network adapter 960 communicates with other modules of electronic device 900 via bus 930. It should be understood that, although... Figure 9 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID (Redundant Arrays of Independent Disks) systems, tape drives, and data backup storage systems.

[0201] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0202] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0203] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is defined only by the appended claims.

Claims

1. A method for recommending musicians, characterized in that, The method includes: Based on a reference high-quality music library, a first target song is determined from a candidate music library, and a first target musician is determined based on the musician information corresponding to the first target song; the reference high-quality music library is constructed from multiple sub-music libraries, and different sub-music libraries contain different types of high-quality songs; the candidate music library refers to a music library published by musicians for recommendation to users; the first target song is a song that possesses the characteristics of high-quality songs in the reference high-quality music library; Determine the quality of the songs corresponding to the first target musician, and select a second target musician from the first target musicians based on the quality of the songs corresponding to the first target musicians; Determine the song recommendation index parameters corresponding to the second target musician, and recommend musicians based on the song recommendation index parameters corresponding to the second target musician; The step of determining the first target song from the candidate music library based on a reference library of high-quality music includes: The target reference song with the highest similarity to the first song to be processed is determined from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library; If the similarity between the first song to be processed and the target reference song meets a preset similarity threshold, then the first song to be processed is taken as the first target song.

2. The method according to claim 1, characterized in that, The step of determining the target reference song with the highest similarity to the first song to be processed from the reference library includes: Based on the representation extraction model, the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library are extracted. The representational features corresponding to the first song to be processed are compared with the representational features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library. Based on the similarity between the first song to be processed and each song in the reference high-quality music library, a target reference song with the highest similarity to the first song to be processed is determined from the reference high-quality music library.

3. The method according to claim 1, characterized in that, Before determining the first target song from the candidate music library based on a reference library of high-quality music, the method further includes: Identify and remove abnormal songs from the reference library of high-quality music.

4. The method according to claim 3, characterized in that, The process of identifying abnormal songs in the referenced premium music library includes: Based on the representation extraction model, the representation features corresponding to each song in the reference high-quality music library are extracted; The representative features of each song in the reference music library are clustered, and songs with representative features outside the class are regarded as abnormal songs.

5. The method according to claim 2 or 4, characterized in that, The representation extraction model is constructed in the following manner: A style tag model for style tag classification is trained based on song tag sample data, resulting in a trained style tag model. The song tag sample data includes style tag data corresponding to various types of songs. By adding an encoding neural network branch for outputting a specified dimension to the already trained style tag model, a Siamese network model is obtained. The Siamese network model is trained based on song fragment sample data to obtain the representation extraction model.

6. The method according to claim 5, characterized in that, The song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs, wherein the song fragments in the positively correlated sample data pairs belong to the same song, and the song fragments in the negatively correlated sample data pairs do not belong to the same song. The Siamese network model is trained based on song fragment sample data to obtain the representation extraction model, including: Based on the song fragment sample data, the Siamese network model is trained using a multi-task learning approach that includes label classification learning and metric learning to obtain the representation extraction model.

7. The method according to claim 1, characterized in that, Determining the song quality corresponding to the first target musician includes: Determine the melody quality score and audio quality score corresponding to the first target musician; Based on the melody quality score and the audio quality score, the song quality corresponding to the first target musician is determined.

8. The method according to claim 7, characterized in that, Determining the melody quality score and audio quality score corresponding to the first target musician includes: Based on the melody information of one or more songs by the first target musician, determine the melody quality score of the first target musician; Based on the audio information of one or more songs by the first target musician, determine the audio quality score of the first target musician.

9. The method according to claim 1, characterized in that, The step of selecting a second target musician from the first target musicians based on the quality of their corresponding songs includes: If the quality of the song corresponding to the first target musician reaches a preset quality threshold, then the first target musician will be used as the second target musician.

10. The method according to claim 1, characterized in that, The song recommendation metrics include style tags and / or popularity scores. The steps for determining the song recommendation metrics corresponding to the second target musician include: Determine the style tags corresponding to one or more songs by the second target musician, and based on the style tags corresponding to one or more songs by the second target musician, determine the style tag corresponding to the second target musician; and / or Determine the popularity score corresponding to one or more songs of the second target musician, and based on the popularity score corresponding to one or more songs of the second target musician, determine the popularity score corresponding to the second target musician.

11. The method according to claim 10, characterized in that, The step of determining the style tags corresponding to one or more songs by the second target musician includes: Based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference music library, the style tag corresponding to the second song to be processed is determined, wherein the second song to be processed is a song by the second target musician.

12. The method according to claim 10, characterized in that, Determining the popularity rating of one or more songs by the second target musician includes: Based on the popularity information corresponding to the second song to be processed, a popularity score is determined for the second song to be processed, wherein the second song to be processed is a song by the second target musician.

13. The method according to claim 10, characterized in that, The step of recommending musicians based on the song recommendation index parameters corresponding to the second target musician includes: Based on the style tags corresponding to the second target musicians, musicians who match the user tags are identified from the second target musicians; Musicians are recommended to users corresponding to the user tags in descending order of popularity ratings corresponding to the musicians that match the user tags.

14. A musician recommendation device, characterized in that, The device includes: The first filtering module is used to determine a first target song from the candidate song library based on the reference high-quality song library, and to determine a first target musician based on the musician information corresponding to the first target song; the first target song is a song that has the characteristics of a high-quality song in the reference high-quality song library; The second filtering module is used to determine the quality of the songs corresponding to the first target musician, and to filter the second target musician from the first target musician based on the quality of the songs corresponding to the first target musician; the reference high-quality music library is constructed from multiple sub-music libraries, and different sub-music libraries contain different types of high-quality songs; the candidate music library refers to the music library published by musicians for recommendation to users; The recommendation module is used to determine the song recommendation index parameters corresponding to the second target musician, and to recommend musicians based on the song recommendation index parameters corresponding to the second target musician; The first filtering module includes: The target reference song determination module is used to determine the target reference song with the highest similarity to the first song to be processed from the reference high-quality music library, wherein the first song to be processed is any song in the candidate music library; The first target song identification module is used to identify the first target song as the first target song if the similarity between the first song to be processed and the target reference song meets a preset similarity threshold.

15. The apparatus according to claim 14, characterized in that, The target reference song determination module is configured as follows: Based on the representation extraction model, the representation features corresponding to the first song to be processed and the representation features corresponding to each song in the reference high-quality music library are extracted. The representational features corresponding to the first song to be processed are compared with the representational features corresponding to each song in the reference high-quality music library to determine the similarity between the first song to be processed and each song in the reference high-quality music library. Based on the similarity between the first song to be processed and each song in the reference high-quality music library, a target reference song with the highest similarity to the first song to be processed is determined from the reference high-quality music library.

16. The apparatus according to claim 14, characterized in that, Before determining the first target song from the candidate music library based on a reference library of high-quality music, the device further includes: The abnormal song removal module is used to identify abnormal songs in the reference high-quality music library and remove them.

17. The apparatus according to claim 16, characterized in that, The abnormal song removal module includes: The representation feature determination module is used to extract the representation features corresponding to each song in the reference high-quality music library based on the representation extraction model. The abnormal song identification module is used to cluster the representational features corresponding to each song in the reference high-quality music library, and identify songs with representational features outside the class as abnormal songs.

18. The apparatus according to claim 15 or 17, characterized in that, The device further includes: a characterization extraction model construction module; The representation extraction model construction module includes: The style tag model training module is used to train a style tag model for style tag classification based on song tag sample data, and obtain the trained style tag model. The song tag sample data includes style tag data corresponding to various types of songs. The Siamese network model building module is used to add a branch of the encoded neural network for outputting a specified dimension to the trained style label model to obtain the Siamese network model. The Siamese network model training module is used to train the Siamese network model based on song fragment sample data to obtain the representation extraction model.

19. The apparatus according to claim 18, characterized in that, The song fragment sample data includes positively correlated sample data pairs and negatively correlated sample data pairs, wherein the song fragments in the positively correlated sample data pairs belong to the same song, and the song fragments in the negatively correlated sample data pairs do not belong to the same song. The twin network model training module includes: Based on the song fragment sample data, the Siamese network model is trained using a multi-task learning approach that includes label classification learning and metric learning to obtain the representation extraction model.

20. The apparatus according to claim 14, characterized in that, The second filtering module includes: The quality scoring module is used to determine the melody quality score and audio quality score corresponding to the first target musician; The quality determination module is used to determine the song quality corresponding to the first target musician based on the melody quality score and the audio quality score.

21. The apparatus according to claim 20, characterized in that, The quality scoring module includes: The melody quality scoring module is used to determine the melody quality score of the first target musician based on the melody information corresponding to one or more songs of the first target musician. The audio quality scoring module is used to determine the audio quality score of the first target musician based on the audio information corresponding to one or more songs of the first target musician.

22. The apparatus according to claim 14, characterized in that, The second filtering module includes: The premium musician identification module is used to designate the first target musician as the second target musician if the quality of the song corresponding to the first target musician reaches a preset quality threshold.

23. The apparatus according to claim 14, characterized in that, The song recommendation metrics include style tags and / or popularity scores. The recommendation module includes: The musician style determination module is used to determine the style tags corresponding to one or more songs of the second target musician, and based on the style tags corresponding to one or more songs of the second target musician, determine the style tag corresponding to the second target musician; and / or The musician popularity determination module is used to determine the popularity score corresponding to one or more songs of the second target musician, and to determine the popularity score corresponding to the second target musician based on the popularity score corresponding to one or more songs of the second target musician.

24. The apparatus according to claim 23, characterized in that, The musician style determination module includes: The song style determination module is used to determine the style tag corresponding to the second song to be processed based on the preset style tag corresponding to the target reference song with the highest similarity to the second song to be processed in the reference boutique music library, wherein the second song to be processed is a song by the second target musician.

25. The apparatus according to claim 23, characterized in that, The musician popularity determination module includes: The song popularity determination module is used to determine the popularity score of the second song to be processed based on the popularity information corresponding to the second song to be processed, wherein the second song to be processed is a song by the second target musician.

26. The apparatus according to claim 23, characterized in that, The recommendation module also includes: The musician style matching module is used to determine the musicians that match the user's tags from the second target musicians based on the style tags corresponding to the second target musicians; The musician recommendation module is used to recommend musicians to users corresponding to the user tags, in descending order of popularity ratings corresponding to the musicians that match the user tags.

27. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 13.

28. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 13.

Images