Song recommendation method and device and storage medium

Abstract

The invention provides a song recommendation method and device and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: determining a target song recommendation scene and a target user account as a song recommendation target; determining the weight of a private expert network corresponding to each song recommendation scene in the plurality of song recommendation scenes in the target song recommendation scene, and inputting song information of songs of historical operation of the target user account into each private expert network, determining user behavior characteristics according to the characteristics output by each private expert network and the weight of each private expert network in the target song recommendation scene; inputting the user behavior features into a scene preference feature extraction model corresponding to the target song recommendation scene to obtain scene preference features corresponding to the target user account in the target song recommendation scene; and based on the scene preference features, determining a song to be recommended in a song library, and recommending the song to be recommended to the target user account. By adopting the method, songs can be recommended individually.

Description

Technical Field

[0001] The present application relates to the field of audio technology, and in particular to a song recommendation method, device and storage medium. Background Art

[0002] Song recommendation is one of the main ways to expose songs. At present, multiple music recommendation sections are usually set up in various music applications. Each music recommendation section corresponds to a music recommendation scenario, such as "Guess you like", "30 songs a day", "Single recommendation" and other recommendation scenarios. In different music recommendation scenarios, the songs recommended to users are usually different. Therefore, how to recommend songs to users according to the music recommendation scenario is a problem that needs to be solved at present. Summary of the invention

[0003] The embodiment of the present application provides a song recommendation method, device and storage medium, which can make personalized song recommendations to users according to different song recommendation scenarios. The technical solution is as follows:

[0004] In a first aspect, a method for song recommendation is provided, the method comprising:

[0005] Determine a target song recommendation scenario and a target user account as a song recommendation target; wherein the target song recommendation scenario is any one of a plurality of preset song recommendation scenarios, and each of the preset song recommendation scenarios corresponds to a private expert network;

[0006] Determine the weight of the private expert network corresponding to each of the multiple song recommendation scenarios in the target song recommendation scenario,

[0007] Inputting song information of songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determining the user behavior features corresponding to the target user account according to the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario;

[0008] Inputting the user behavior feature into a scene preference feature extraction model corresponding to the target song recommendation scene to obtain the scene preference feature corresponding to the target user account in the target song recommendation scene;

[0009] Based on the scene preference characteristics corresponding to the target user account in the target song recommendation scene, the songs to be recommended are determined in the music library, and the songs to be recommended are recommended to the target user account.

[0010] In a second aspect, a device for song recommendation is provided, the device comprising:

[0011] A determination module is used to determine a target song recommendation scenario and a target user account as a song recommendation target; wherein the target song recommendation scenario is any one of a plurality of preset song recommendation scenarios, and each of the preset song recommendation scenarios corresponds to a private expert network; the weight of the private expert network corresponding to each of the plurality of song recommendation scenarios in the target song recommendation scenario is determined, song information of songs historically operated by the target user account is input into the private expert network corresponding to each song recommendation scenario, and user behavior features corresponding to the target user account are determined based on features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario; the user behavior features are input into a scene preference feature extraction model corresponding to the target song recommendation scenario to obtain scene preference features corresponding to the target user account in the target song recommendation scenario;

[0012] The recommendation module is used to determine the songs to be recommended in the music library based on the scene preference characteristics corresponding to the target user account in the target song recommendation scene, and recommend the songs to be recommended to the target user account.

[0013] According to a third aspect, a computing device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the operations performed by the song recommendation method as described in the first aspect and any possible implementation of the first aspect.

[0014] In a fourth aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the instruction is loaded and executed by a processor to implement the operations performed by the song recommendation method as described in the first aspect and any possible implementation of the first aspect.

[0015] In a fifth aspect, a computer program product is provided, wherein at least one instruction is stored in the computer program product, and the instruction is loaded and executed by a processor to implement the operations performed by the song recommendation method as described in the first aspect and any possible implementation of the first aspect.

[0016] The beneficial effects of the technical solution provided by this application are:

[0017] In the song recommendation method provided in the embodiment of the present application, a corresponding private expert network and a scene preference feature extraction model are introduced for each music recommendation scenario. In the target music recommendation scenario, when recommending songs to users, the corresponding private expert network is first used to extract features of the songs historically operated by the user to obtain the user behavior features corresponding to the user, and then the user behavior features are input into the scene preference feature extraction model corresponding to the target song recommendation scenario to obtain the scene preference features corresponding to the user in the target song recommendation scenario. The scene preference features obtained in this way can reflect the user's preference for songs in the target song recommendation scenario. Based on this, matching songs to be recommended are selected from the music library as songs recommended to users in the target song recommendation scenario. These songs meet the user's song preferences in the target song recommendation scenario, thereby realizing personalized song recommendations for different users in different song recommendation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a flow chart of a song recommendation method provided by an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of an expert network provided in an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of an expert network provided in an embodiment of the present application;

[0022] Figure 4 is a schematic diagram of an expert network provided in an embodiment of the present application;

[0023] Figure 5 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0024] Figure 6 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0025] Figure 7 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0026] Figure 8 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0027] Fig. 9 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0028] Fig.10 is a schematic diagram of a scene preference feature extraction model provided in an embodiment of the present application;

[0029] Fig.11 is a schematic diagram of the structure of a song recommendation device provided in an embodiment of the present application;

[0030] Fig.12 is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0031] Fig.13 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0033] An embodiment of the present application provides a method for song recommendation, which can be implemented by a computing device. The computing device can be a server, and the server can be a single server, a server cluster, etc.

[0034] The implementation scenario of the song recommendation method provided in the embodiment of the present application is described below. A music application can be installed in the terminal, and multiple song recommendation modules can be displayed in the interface of the music application, such as "Guess You Like", "30 Songs a Day", "Single Recommendation", etc. Each song recommendation module corresponds to a song recommendation scenario. The user can log in to the user account in the music application, and then the user can operate any song recommendation module displayed in the interface of the music application, and trigger the terminal to send a song recommendation request for the song recommendation scenario corresponding to the song recommendation module to the server. After the server receives the song recommendation request, it can use the song recommendation method provided in the embodiment of the present application to return the songs recommended to the user in the song recommendation scenario to the terminal.

[0035] The following is an explanation of the song recommendation method provided in the embodiment of the present application. The method can be implemented by a computing device. Figure 1 , the processing flow of the method may include the following steps:

[0036] Step 101: Determine a target song recommendation scenario and a target user account as a song recommendation target.

[0037] The target song recommendation scenario is any preset song recommendation scenario among multiple preset song recommendation scenarios, and the target user account is an account registered by any user in the music application.

[0038] In implementation, a music application may be installed in the terminal, and the user may log in to the target user account in the music application. The user may select any music recommendation module displayed in the interface of the music application, and the song recommendation scene corresponding to the selected music recommendation module is the target song recommendation scene. After the terminal detects the selection operation of the music recommendation module, it sends a song recommendation request to the server, wherein the song recommendation request carries the identifier of the target song recommendation scene corresponding to the song recommendation module.

[0039] After receiving the song recommendation request, the server obtains the songs to be recommended corresponding to the target user account in the target song recommendation scenario. Here, obtaining the songs to be recommended may refer to: after receiving the song recommendation request, executing the song recommendation method provided in the embodiment of the present application to obtain the songs to be recommended; or it may refer to: directly reading the songs to be recommended corresponding to the target user account in the target song recommendation scenario in the storage medium. In this case, the server may have executed the song recommendation method provided in the embodiment of the present application to obtain the songs to be recommended before receiving the song recommendation request. The triggering of the execution may be periodic. For example, the song recommendation method provided in the embodiment of the present application is executed once at 00:00 every day to obtain the songs to be recommended corresponding to each user account in each song recommendation scenario. Of course, the timing of obtaining the songs to be recommended in different song recommendation scenarios may be different or the same, and the embodiment of the present application does not limit this.

[0040] Step 102: Determine the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario among the multiple song recommendation scenarios.

[0041] Among them, the expert network can be constructed by DNN (Deep Neural Networks) or other forms of neural network models. Each song recommendation scenario corresponds to an expert network. In order to distinguish it from the shared expert network, the expert network corresponding to the song recommendation scenario can be called a private expert network. It can be understood that a private expert network refers to an expert network that is good at feature extraction in a certain song recommendation scenario.

[0042] In implementation, each song recommendation scenario can correspond to a private expert network, and each private expert network can correspond to a gating network. Figure 2 , assuming that there are three song recommendation scenarios, corresponding to private expert network 1, private expert network 2 and private expert network 3 respectively, among which private expert network 1 corresponds to gated network 1, private expert network 2 corresponds to gated network 2, and private expert network 3 corresponds to gated network 3.

[0043] In a possible implementation, the server may input the identifier of the target song recommendation scenario into the gating network corresponding to each private expert network, and each gating network outputs the weight of the corresponding private expert network in the target song recommendation scenario.

[0044] In a possible implementation, the weight of each private expert network in the target song recommendation scenario can also be determined in combination with the attribute information of the target user. Specifically, the identifier of the target song recommendation scenario and the attribute information of the target user can be input into the gating network corresponding to each private expert network, and each gating network outputs the weight of the corresponding private expert network in the target song recommendation scenario.

[0045] It should be noted that the "input" here can be direct input or indirect input. Indirect input can be input after converting the identification and attribute information into a vector of a specified dimension. The embodiments of the present application do not limit this.

[0046] How to obtain the attribute information of the target user: the server can obtain the attribute information of the target user corresponding to the target user account from the attribute information of the user filled in when each user account is registered. The attribute information of the user may include the user's age, gender, city level, education level, job type, etc.

[0047] Step 103: input the song information of the songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determine the user behavior characteristics corresponding to the target user account based on the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario.

[0048] The server can obtain song information of songs operated historically by the target user account. The historical operation can be any operation such as recent collection, comment, and completion. The recent operation can be the past week, the past month, etc., which can be configured by relevant personnel according to actual needs. The song information includes at least one of the following information: the song title, the time difference between the time of the song operated historically by the target user account and the current time, the song genre, the album to which the song belongs, the singer, etc.

[0049] In a possible implementation, the server inputs the song information of the songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and obtains the first feature output by each private expert network. For the first feature output by each private expert network, the first feature is multiplied by the weight corresponding to the private expert network to obtain the second feature. Then, the second features corresponding to each private expert network are concatenated to obtain the user behavior feature corresponding to the target user account.

[0050] Song information can be input into the private expert network in the form of a sequence. Specifically, the song titles can be composed of a first sequence, the time difference between the target user account's historical song operations and the current time can be composed of a second sequence, the song genres can be composed of a third sequence, the first sequence, the second sequence, and the third sequence can be composed of a fourth sequence, and the fourth sequence can be input into the private expert network.

[0051] See also Figure 3 , the song information of the songs historically operated by the target user account is input into private expert network 1, private expert network 2 and private expert network 3 respectively, private expert network 1 outputs first feature 1, private expert network 2 outputs first feature 2, and private expert network 3 outputs first feature 3. The identifier of the target song recommendation scene and the attribute information of the target user are input into gated network 1, gated network 2 and gated network 3, gated network 1 outputs weight 1, gated network 2 outputs weight 2, and gated network 3 outputs weight 3. Weight 1 is multiplied by first feature 1 to obtain second feature 1, weight 2 is multiplied by first feature 2 to obtain second feature 2, and weight 3 is multiplied by first feature 3 to obtain second feature 3. The vectors corresponding to second feature 1, second feature 2 and second feature 3 are concatenated to obtain the user behavior features corresponding to the target user account.

[0052] In another possible implementation, when extracting user behavior features, in addition to the private expert network corresponding to each song recommendation scenario, a shared expert network can also be introduced to improve the universality and stability of the entire system. Accordingly, the processing of the above step 103 can be as follows:

[0053] The server inputs the song information of the songs historically operated by the target user account into the private expert network and the shared expert network corresponding to each song recommendation scenario, and obtains the first feature output by each private expert network and the third feature output by the shared expert. For the first feature output by each private expert network, the first feature and the third feature are weighted based on the weight corresponding to the private expert network to obtain the fourth feature. Then, the fourth feature corresponding to each private expert network is spliced ​​to obtain the user behavior feature corresponding to the target user account.

[0054] In this case, the weight corresponding to each private expert network may include a first weight and a second weight. Accordingly, based on the weight corresponding to the private expert network, the weighted processing of the first feature and the third feature may be: based on the first weight and the second weight of the private expert network in the target song recommendation scenario, the first feature and the third feature are weighted and summed to obtain the fourth feature. Specifically, the first weight and the first feature may be multiplied, the second weight and the third feature may be multiplied, and the obtained products may be added to obtain the fourth feature.

[0055] The following is an example to illustrate:

[0056] See also Figure 4 ,exist Figure 2 Based on the above, a shared expert network is introduced. The song information of the songs operated historically by the target user account is input into private expert network 1, private expert network 2, private expert network 3 and shared expert network respectively. Private expert network 1 outputs the first feature 1, private expert network 2 outputs the first feature 2, private expert network 3 outputs the first feature 3, and the shared expert network outputs the third feature. Gated network 1 outputs the first weight 1 and the second weight 1, gated network 2 outputs the first weight 2 and the second weight 2, and gated network 3 outputs the first weight 3 and the second weight 3. The first weight 1 is multiplied by the first feature 1, the second weight 1 is multiplied by the third feature, and the products are added to obtain the fourth feature 1. The first weight 2 is multiplied by the first feature 2, the second weight 2 is multiplied by the third feature, and the products are added to obtain the fourth feature 2. The first weight 3 is multiplied by the first feature 3, the second weight 3 is multiplied by the third feature, and the products are added to obtain the fourth feature 3. Then, the vectors corresponding to the fourth feature 1, the fourth feature 2, and the fourth feature 3 are concatenated to obtain the user behavior features corresponding to the target user account.

[0057] In a possible implementation, the basic features of the target user account and the features output by the expert network can be combined to obtain the user behavior features, wherein the basic features of the target user account are used to indicate the activity level of the user account in the above music application.

[0058] Correspondingly, after obtaining the second feature as described above, the second feature corresponding to each private expert network and the basic feature of the target user account can be concatenated to obtain the user behavior feature corresponding to the target user account. Figure 5 As shown, in Figure 3 On this basis, the vectors corresponding to the second feature 1, the second feature 2, the second feature 3 and the user behavior features corresponding to the target user account are concatenated to obtain the user behavior features corresponding to the target user account.

[0059] After obtaining the fourth feature as described above, the fourth feature corresponding to each private expert network and the basic feature of the target user account can be concatenated to obtain the user behavior feature corresponding to the target user account. Figure 6 As shown, in Figure 4 On this basis, the vectors corresponding to the fourth feature 1, the fourth feature 2, the fourth feature 3 and the user behavior features corresponding to the target user account are concatenated to obtain the user behavior features corresponding to the target user account.

[0060] Step 104: Input the user behavior characteristics into the scene preference feature extraction model corresponding to the target song recommendation scene to obtain the scene preference characteristics corresponding to the target user account in the target song recommendation scene.

[0061] Among them, the scene preference feature extraction model can be a DNN model.

[0062] In implementation, each song recommendation scenario may correspond to a scenario preference feature extraction model. The server may input the user behavior features into each scenario preference feature extraction model respectively, and obtain the scenario preference features output by the scenario preference feature extraction model corresponding to the target song recommendation scenario as the scenario preference features corresponding to the target user account in the target song recommendation scenario. Alternatively, the server may input the user behavior features only into the scenario preference feature extraction model corresponding to the target song recommendation scenario, and obtain the scenario preference features output by the scenario preference feature extraction model corresponding to the target song recommendation scenario as the scenario preference features corresponding to the target user account in the target song recommendation scenario.

[0063] See also Figure 7 , assuming that there are three song recommendation scenarios, corresponding to scenario preference feature extraction model 1, scenario preference feature extraction model 2 and scenario preference feature extraction model 3 respectively, and the scenario preference feature extraction model corresponding to the target song recommendation scenario is scenario preference feature extraction model 2. The server can Figure 3 or Figure 4 The output user behavior features are input into the scene preference feature extraction model 2 to obtain the scene preference features output by the scene preference feature extraction model 2 as the scene preference features corresponding to the target user account in the target song recommendation scenario.

[0064] In a possible implementation, when extracting scene preference features, in addition to the scene preference feature extraction model corresponding to each song recommendation scene, a shared scene preference feature extraction model can also be introduced to improve the universality and stability of the entire system. Accordingly, the processing of the above step 104 can be as follows:

[0065] The user behavior feature is input into the scene preference feature extraction model corresponding to each song recommendation scene to obtain the fifth feature, and the user behavior feature is input into the shared scene preference feature extraction model to obtain the sixth feature. Based on the fifth and sixth features, the scene preference feature corresponding to the target user account in the target song recommendation scene is obtained.

[0066] In implementation, based on the fifth feature and the sixth feature, the process of obtaining the scene preference feature corresponding to the target user account in the target song recommendation scene may be: performing feature fusion on the fifth feature and the sixth feature to obtain the scene preference feature corresponding to the target user account in the target song recommendation scene. Feature fusion may be feature concatenation.

[0067] See also Figure 8 ,exist Figure 7 Based on the above, a shared scene preference feature extraction model is introduced. The server can input the user behavior feature into the scene preference feature extraction model 2 to obtain the fifth feature, and input the user behavior feature into the shared scene preference feature extraction model to obtain the sixth feature. Based on the fifth feature and the sixth feature, the scene preference feature corresponding to the target user account in the target song recommendation scenario is obtained.

[0068] In another possible implementation, for each song recommendation scene, a scene preference feature extraction model corresponding to different activity levels can be set, wherein the activity level is used to characterize the frequency or activity of the user's operation of the song. Accordingly, the processing of the above step 104 can be as follows:

[0069] Determine the activity level of the target user account based on the number of songs historically operated by the target user account. Determine the scene preference feature extraction model corresponding to the activity level of the target user account in at least one scene preference feature extraction model corresponding to the target song recommendation scene. Determine the scene preference features corresponding to the target user account in the target song recommendation scene based on the user behavior features and the scene preference feature extraction model corresponding to the activity level of the target user account.

[0070] In implementation, the activity level of the target user account is determined based on the number of songs historically operated by the target user account and the range of the number of songs corresponding to each activity level. For example, two activity levels are set, the number of songs corresponding to the first activity level ranges from 0 to 100, and the number of songs corresponding to the second activity level ranges from more than 100. The number of songs historically operated by the target user account is 90, then the activity level of the target user account is determined to be the first activity level.

[0071] Then, in the scene preference feature extraction models corresponding to each activity level in the target song recommendation scenario, the scene preference feature extraction model corresponding to the activity level of the target user account is determined. The user behavior features are input into the scene preference feature extraction model corresponding to the activity level of the target user account to obtain the scene preference features corresponding to the target user account in the target song recommendation scenario.

[0072] See also Fig. 9 ,Assume that there are two activity levels and three song recommendation scenarios, then each song recommendation scenario corresponds to two scene preference feature extraction models corresponding to the two activity levels, such as Fig. 9The first scene preference feature extraction model 1 is the scene preference feature extraction model corresponding to the first activity level in song recommendation scenario 1, and the second scene preference feature extraction model 1 is the scene preference feature extraction model corresponding to the second activity level in song recommendation scenario 1; the first scene preference feature extraction model 2 is the scene preference feature extraction model corresponding to the first activity level in song recommendation scenario 2, and the second scene preference feature extraction model 2 is the scene preference feature extraction model corresponding to the second activity level in song recommendation scenario 2; the first scene preference feature extraction model 3 is the scene preference feature extraction model corresponding to the first activity level in song recommendation scenario 3, and the second scene preference feature extraction model 3 is the scene preference feature extraction model corresponding to the second activity level in song recommendation scenario 3.

[0073] It is determined that the activity level of the target user account is the first activity level, and the target song recommendation scenario is song recommendation scenario 2, then the user behavior characteristics are input into the first scenario preference feature extraction model 2 to obtain the scenario preference characteristics corresponding to the target user account in the target song recommendation scenario.

[0074] In another possible implementation, a shared scene preference feature extraction model may be provided, and a scene preference feature extraction model may be provided for each song recommendation scene according to the activity level. Accordingly, the processing of the above step 104 may be as follows:

[0075] The activity level of the target user account is determined according to the number of songs historically operated by the target user account. The scene preference feature extraction model corresponding to the activity level of the target user account is determined in at least one scene preference feature extraction model corresponding to the target song recommendation scene.

[0076] The user behavior characteristics are input into the shared scene preference feature extraction model to obtain the sixth feature. The user behavior characteristics are input into the scene preference feature extraction model corresponding to the activity level of the target user account to obtain the seventh feature. Based on the sixth and seventh features, the scene preference characteristics corresponding to the target user account in the target song recommendation scenario are obtained.

[0077] See also Fig.10 ,Will Figure 8 and Fig. 9 Combined, it is determined that the activity level of the target user account is the first activity level, and the target song recommendation scenario is song recommendation scenario 2, then the user behavior feature is input into the first scenario preference feature extraction model 2 to obtain the seventh feature, and the user behavior feature is input into the shared scenario preference feature extraction model to obtain the sixth feature. Based on the sixth and seventh features, the scenario preference feature corresponding to the target user account in the target song recommendation scenario is obtained.

[0078] In the embodiments of the present application, Figure 6and Fig.10 The combined model builds different private expert networks at the bottom layer to capture user behavior characteristics in different music recommendation scenarios. The top layer draws on the star structure to jointly model users with different levels of activity in different music recommendation scenarios. This ensures that the model can capture users' song preferences according to different music recommendation scenarios, and can separately model users with different levels of activity in the same music recommendation scenario, so that inactive users are not led astray by active users. In this way, both active users and inactive users in different music recommendation scenarios can be well represented by the model, ultimately improving the online performance of the model and enabling the model to better utilize data across music recommendation scenarios to recall recommendation results relevant to users.

[0079] Step 105: Based on the scene preference characteristics corresponding to the target user account in the target song recommendation scene, determine the songs to be recommended in the music library, and recommend the songs to be recommended to the target user account.

[0080] In implementation, the server can extract song features of songs in the music library in advance through a song feature extraction model. Specifically, for each song, the song attribute information of the song is input into the song feature extraction model to obtain song features. The song attribute information includes information representing the basic attributes of the song, such as the album, singer, release year, language, etc., as well as information that can measure the popularity of the song, such as the number of plays, number of completed plays, and number of shares.

[0081] In addition, the above-mentioned song feature extraction model, the above-mentioned private expert network, shared expert network, scene preference feature extraction model, shared scene preference feature extraction model, etc. are jointly trained, wherein the song feature extraction model serves as the item side model in the twin-tower model, and the private expert network, shared expert network, scene preference feature extraction model, shared scene preference feature extraction model, etc. serve as the user side models in the twin-tower model. The specific training method is similar to the training method of the twin-tower model, which will not be repeated here.

[0082] After obtaining the scene preference features corresponding to the target user account in the target song recommendation scene in the above step 105, the similarity between the scene preference features and the song features of the songs in the music library can be calculated to determine the songs that meet the similarity condition. The songs that meet the similarity condition can be songs with a similarity threshold greater than a threshold, or a preset number of songs with the highest similarity.

[0083] After receiving the song recommendation request, the server returns the songs that meet the similarity condition as the songs to be recommended corresponding to the target user account in the target song recommendation scenario to the terminal.

[0084] In the song recommendation method provided in the embodiment of the present application, a corresponding private expert network and a scene preference feature extraction model are introduced for each music recommendation scenario. In the target music recommendation scenario, when recommending songs to users, the corresponding private expert network is first used to extract features of the songs historically operated by the user to obtain the user behavior features corresponding to the user, and then the user behavior features are input into the scene preference feature extraction model corresponding to the target song recommendation scenario to obtain the scene preference features corresponding to the user in the target song recommendation scenario. The scene preference features obtained in this way can reflect the user's preference for songs in the target song recommendation scenario. Based on this, matching songs to be recommended are selected from the music library as songs recommended to users in the target song recommendation scenario. These songs meet the user's song preferences in the target song recommendation scenario, thereby realizing personalized song recommendations for different users in different song recommendation scenarios.

[0085] The technical solution provided in the embodiment of the present application has strong scalability. The model structure (such as the number of private expert networks, the number of scene preference feature extraction models, etc.) can be designed according to the different numbers of music recommendation scenarios and different population segmentation standards. Ultimately, personalized song recommendations can be achieved for different users in different song recommendation scenarios.

[0086] Based on the same technical concept, the embodiment of the present application also provides a device for song recommendation, which can be a computing device, such as Fig.11 As shown, the device includes a determination module 920 and a recommendation module 930, wherein:

[0087] Determination module 920, used to determine a target song recommendation scenario and a target user account as a song recommendation target; wherein the target song recommendation scenario is any preset song recommendation scenario among a plurality of preset song recommendation scenarios, and each of the preset song recommendation scenarios corresponds to a private expert network; determine the weight of the private expert network corresponding to each song recommendation scenario among the plurality of song recommendation scenarios in the target song recommendation scenario, input the song information of the songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determine the user behavior feature corresponding to the target user account according to the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario; input the user behavior feature into the scene preference feature extraction model corresponding to the target song recommendation scenario to obtain the scene preference feature corresponding to the target user account in the target song recommendation scenario;

[0088] The recommendation module 930 is used to determine the songs to be recommended in the music library based on the scene preference characteristics corresponding to the target user account in the target song recommendation scene, and recommend the songs to be recommended to the target user account.

[0089] In a possible implementation, the song information includes at least one of the title of the song, the time difference between the time when the target user account historically operated the song and the current time, and the genre of the song.

[0090] In the song recommendation method provided in the embodiment of the present application, a corresponding private expert network and a scene preference feature extraction model are introduced for each music recommendation scenario. In the target music recommendation scenario, when recommending songs to users, the corresponding private expert network is first used to extract features of the songs historically operated by the user to obtain the user behavior features corresponding to the user, and then the user behavior features are input into the scene preference feature extraction model corresponding to the target song recommendation scenario to obtain the scene preference features corresponding to the user in the target song recommendation scenario. The scene preference features obtained in this way can reflect the user's preference for songs in the target song recommendation scenario. Based on this, matching songs to be recommended are selected from the music library as songs recommended to users in the target song recommendation scenario. These songs meet the user's song preferences in the target song recommendation scenario, thereby realizing personalized song recommendations for different users in different song recommendation scenarios.

[0091] It should be noted that: the song recommendation device provided in the above embodiment only uses the division of the above functional modules as an example when recommending songs. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computing device can be divided into different functional modules to complete all or part of the functions described above. In addition, the song recommendation device provided in the above embodiment and the song recommendation method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0092] Fig.12 The structure block diagram of a terminal 600 provided by an exemplary embodiment of the present application is shown. The terminal 600 may be a portable mobile terminal, such as a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer or a desktop computer. The terminal 600 may also be called a user device, a portable terminal, a laptop terminal, a desktop terminal or other names.

[0093] Typically, the terminal 600 includes a processor 601 and a memory 602 .

[0094] The processor 601 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 601 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 601 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 601 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 601 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0095] The memory 602 may include one or more computer-readable storage media, which may be non-transitory. The memory 602 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices.

[0096] In some embodiments, the terminal 600 may further optionally include: a peripheral device interface 603 and at least one peripheral device. The processor 601, the memory 602 and the peripheral device interface 603 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 603 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 604, a display screen 605, a camera assembly 606, an audio circuit 607, a positioning assembly 608 and a power supply 609.

[0097] In some embodiments, the terminal 600 further includes one or more sensors 610 , including but not limited to: an acceleration sensor 611 , a gyroscope sensor 612 , a pressure sensor 613 , a fingerprint sensor 614 , an optical sensor 615 , and a proximity sensor 616 .

[0098] Those skilled in the art will understand that Figure 7 The structure shown in the figure does not constitute a limitation on the terminal 600, and the terminal 600 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0099] Fig.13 It is a schematic diagram of the structure of a computing device provided in an embodiment of the present application. The computing device 1000 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 1001 and one or more memories 1002, wherein the memory 1002 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 1001 to implement the methods provided in the above-mentioned various method embodiments. Of course, the computing device may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the computing device may also include other components for implementing device functions, which will not be described in detail here.

[0100] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, and the instructions can be executed by a processor in a terminal to complete the method in the above embodiment. The computer-readable storage medium can be non-transitory. For example, the computer-readable storage medium can be a ROM (Read-Only Memory), a RAM (Random Access Memory), a CD-ROM (Compact Disc Read-Only Memory), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0101] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. For example, the user's attribute information and song information of the songs that the user has historically operated on in this application are all obtained with full authorization.

[0102] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0103] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A song recommendation method, characterized in that: The method comprises: Determine a target song recommendation scenario and a target user account as a song recommendation target; wherein the target song recommendation scenario is any one of a plurality of preset song recommendation scenarios, and each of the preset song recommendation scenarios corresponds to a private expert network; Determine the weight of the private expert network corresponding to each of the multiple song recommendation scenarios in the target song recommendation scenario, Inputting song information of songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determining the user behavior features corresponding to the target user account according to the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario; Inputting the user behavior feature into a scene preference feature extraction model corresponding to the target song recommendation scene to obtain the scene preference feature corresponding to the target user account in the target song recommendation scene; Based on the scene preference characteristics corresponding to the target user account in the target song recommendation scene, the songs to be recommended are determined in the music library, and the songs to be recommended are recommended to the target user account.

2. The method according to claim 1, characterized in that The song information includes at least one of the title of the song, the time difference between the time when the target user account historically operated the song and the current time, and the genre of the song.

3. The method according to claim 1, characterized in that The step of inputting song information of songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determining the user behavior features corresponding to the target user account according to the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario, includes: Inputting the song information of the songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, respectively, to obtain the first feature output by each private expert network; Multiplying the first feature output by each of the private expert networks and the weight of the private expert network in the target song recommendation scenario to obtain a second feature corresponding to each of the private expert networks; Based on the second feature corresponding to each of the private expert networks, a user behavior feature corresponding to the target user account is obtained.

4. The method according to claim 3, characterized in that The obtaining of the user behavior feature corresponding to the target user account based on the second feature corresponding to each of the private expert networks includes: Based on the second feature corresponding to each of the private expert networks and the attribute information corresponding to the target user account, a user behavior feature corresponding to the target user account is obtained.

5. The method according to claim 1, characterized in that The step of inputting song information of songs historically operated by the target user account into the private expert network corresponding to each song recommendation scenario, and determining the user behavior features corresponding to the target user account according to the features output by the private expert network corresponding to each song recommendation scenario and the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario, includes: Inputting the song information of the songs historically operated by the target user account into the private expert network and the shared expert network corresponding to each song recommendation scenario respectively, and obtaining the first feature output by each private expert network and the third feature output by the shared expert network; Based on the weight of each of the private expert networks in the target song recommendation scenario, weighting the first feature and the third feature to obtain a fourth feature corresponding to each of the private expert networks; Based on the fourth feature corresponding to each of the private expert networks, a user behavior feature corresponding to the target user account is obtained.

6. The method according to claim 5, characterized in that The determining of the weight of the private expert network corresponding to each song recommendation scenario in the target song recommendation scenario includes: Input the song recommendation scenario identifier of the target song recommendation scenario into the gating network corresponding to each private expert network to obtain the first weight and the second weight of each expert network in the target song recommendation scenario; The first feature and the third feature are weighted based on the weight of each private expert network in the target song recommendation scenario to obtain a fourth feature corresponding to each private expert network, including: Based on the first weight and the second weight of each of the private expert networks in the target song recommendation scenario, the first feature and the third feature are weightedly summed to obtain a fourth feature.

7. The method according to any one of claims 1 to 6, characterized in that The step of inputting the user behavior feature into a scene preference feature extraction model corresponding to the target song recommendation scene to obtain the scene preference feature corresponding to the target user account in the target song recommendation scene includes: Inputting the user behavior feature into the scene preference feature extraction model corresponding to each song recommendation scene to obtain a fifth feature; Inputting the user behavior feature into a shared scene preference feature extraction model to obtain a sixth feature; Based on the fifth feature and the sixth feature, the scene preference feature corresponding to the target user account in the target song recommendation scene is obtained.

8. The method according to any one of claims 1 to 6, characterized in that The step of inputting the user behavior feature into a scene preference feature extraction model corresponding to the target song recommendation scene to obtain the scene preference feature corresponding to the target user account in the target song recommendation scene includes: Determining the activity level of the target user account according to the number of songs historically operated by the target user account; Determine, in at least one scene preference feature extraction model corresponding to the target song recommendation scene, a scene preference feature extraction model corresponding to the activity level of the target user account; Based on the user behavior characteristics and the scene preference feature extraction model corresponding to the activity level of the target user account, the scene preference features corresponding to the target user account in the target song recommendation scenario are determined.

9. A computing device, characterized in that The computing device includes a processor and a memory, wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the operations performed by the song recommendation method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, which is loaded and executed by the processor to implement the operation performed by the song recommendation method as described in any one of claims 1 to 8.

11. A computer program product, characterized in that The computer program product stores at least one instruction, which is loaded and executed by a processor to implement the operations performed by the song recommendation method as described in any one of claims 1 to 8.

Images