New Playlist Recommendation Method, Its Device, Equipment, Medium, and Product

Abstract

This application relates to the technical field of music information retrieval, and discloses a method and device, equipment, medium, and product for recommending new playlists. The sorting method includes: responding to a user's recommendation request, obtaining the user feature vector according to the user's historical behavior data; calculating the similarity between the user feature vector and the pre-collected playlist feature vectors of new playlists in the music library to obtain corresponding similarity values; distributing multiple new playlists whose similarity values meet the preset conditions to the user, and correspondingly obtaining the user behavior data corresponding to the access of the multiple new playlists by the user; screening out target new playlists according to the access popularity of each new playlist statistically calculated from the user behavior data corresponding to the access of all new playlists in the music library by different users, and constructing the target new playlists into a recommended list. This application realizes screening out new playlists according to user behavior for playlist recommendation.

Description

Technical Field

[0001] The present application relates to the technical field of music information retrieval, and in particular to a method for recommending new playlists and its corresponding device, computer device, computer-readable storage medium, and computer program product. Background Art

[0002] With the improvement of the quality of people's spiritual and cultural life, users' enthusiasm for creating playlists is getting higher and higher, producing many high-quality UGC playlists (i.e., user-generated playlists). For each online music platform, a large number of user-generated playlists need to be reviewed every day, and then high-quality user-generated playlists are selected for playlist recommendation. Generally, this work is mainly judged by professionals in each online music platform completely relying on standards and their personal aesthetics, and the output is very limited. Moreover, due to the large amount of work and the need for a relatively large number of relevant personnel, the following specific problems will inevitably exist:

[0003] 1. The cost is high, and professional labor costs are required for a long time to carry out the review work. Moreover, as the number of users increases, the number of required professionals also gradually increases, and it is very likely to evolve into a situation where professional human resources are in short supply and the review work cannot be carried out.

[0004] 2. Different professionals have different aesthetics and trade-offs when reviewing, which is likely to lead to uneven quality of the playlists reviewed and stored in the library. Some high-quality playlists may also be eliminated, and fairness cannot be guaranteed.

[0005] Generally speaking, for the recommendation of new playlists in the prior art, it is difficult to unify the review standards, resulting in quality problems in the reviewed playlists. On the other hand, the manpower required for reviewing playlists is gradually lacking. Therefore, the applicant has made corresponding explorations. Summary of the Invention

[0006] The primary objective of the present application is to solve at least one of the above problems, and to provide a method for recommending new playlists and its corresponding device, computer device, computer-readable storage medium, and computer program product.

[0007] To meet the various objectives of the present application, the following technical solutions are adopted:

[0008] A method for recommending new playlists provided to meet one of the objectives of the present application includes the following steps:

[0009] In response to a user's recommendation request, obtain the user feature vector according to the user's historical behavior data, and the user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data;

[0010] Calculate the similarity between the user feature vector and the pre-collected playlist feature vector of the newly added playlists in the music library to obtain the corresponding similarity value; the playlist feature vector is the feature representation of the deep semantic information of the graphic and text information of the corresponding newly added playlist.

[0011] Distribute multiple newly added playlists whose similarity values meet the preset conditions to the user, and correspondingly obtain the user behavior data corresponding to the access of the multiple newly added playlists by the user.

[0012] Screen out the target newly added playlists according to the access popularity of each newly added playlist statistically obtained from the user behavior data corresponding to the access of all newly added playlists in the music library by different users, and construct the target newly added playlists into a recommended list.

[0013] In a further embodiment, obtaining the user feature vector according to the historical behavior data of the user includes the following steps:

[0014] Obtain the mapping relationship data between the unique feature information of the playlists and songs accessed by the user and the behavior types implemented by the user from the historical behavior data of the user, and the behavior types include any one or more of click, play, favorite, download, and share;

[0015] Encode according to the mapping relationship data to obtain a user encoding vector;

[0016] Adopt a pre-trained and converged user feature extraction model to extract features according to the user encoding vector to obtain the corresponding user feature vector.

[0017] In a further embodiment, distributing multiple newly added playlists whose similarity values meet the preset conditions to the user and correspondingly obtaining the user behavior data corresponding to the access of the multiple newly added playlists by the user includes the following steps:

[0018] Detect whether the total historical distribution volume corresponding to the multiple newly added playlists reaches the preset threshold. When it reaches, stop distributing the corresponding newly added playlists; otherwise, distribute the corresponding newly added playlists to the user;

[0019] Monitor the user behavior data implemented by the user based on the distributed newly added playlists;

[0020] In response to the user behavior data submitted by the user, detect and obtain the user behavior data for the distributed newly added playlists.

[0021] In a further embodiment, screening out the target newly added playlists according to the access popularity of each newly added playlist statistically obtained from the user behavior data corresponding to the access of all newly added playlists in the music library by different users and constructing the target newly added playlists into a recommended list includes the following steps:

[0022] For all newly added playlists in the music library, based on the user behavior data generated by being accessed by different users, count the access volume corresponding to different behavior types imposed on each newly added playlist, where the behavior types include any one or more of click, play, favorite, download, and share;

[0023] For each newly added playlist, perform a weighted sum of the access volumes of different behavior types to obtain the access popularity corresponding to each newly added playlist;

[0024] Judge whether the access popularity of each newly added playlist exceeds a preset threshold, and determine the newly added playlists with access popularity exceeding the preset threshold as target newly added playlists;

[0025] Create a recommendation list and add the target newly added playlists to this recommendation list.

[0026] In a further embodiment, after the step of constructing the target newly added playlists into the recommendation list, the following steps are included:

[0027] In response to a user's playlist recommendation request, according to the user feature vector of this user, match one or more preferred newly added playlists from the newly added playlists in the recommendation list whose playlist feature vectors are similar to the user feature vector, and push the preferred newly added playlists to the corresponding users.

[0028] In a further embodiment, before responding to a user recommendation request, the following steps are included:

[0029] Adopt a predicted data set to train a dual - tower model for extracting the user feature vector and the playlist feature vector until it reaches a convergent state; wherein, the data set includes multiple data samples, each data sample represents the mapping relationship data between the unique feature information of the playlists and songs accessed by a corresponding user and the behavior types implemented by this user, and the playlist feature data is the graphic and text information of the playlists accessed by the corresponding user;

[0030] Call the dual - tower model that has been trained to a convergent state, and for any playlist in the music library, extract its corresponding playlist feature vector according to the graphic and text information of this playlist.

[0031] A newly added playlist recommendation device provided to meet one of the purposes of the present application includes: a user processing module, a similarity matching module, a playlist distribution module, and a playlist screening module. Among them, the user processing module is used to respond to a user recommendation request and obtain a user feature vector according to the historical behavior data of the user. The user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data; the similarity matching module is used to calculate the similarity between the user feature vector and the pre-collected playlist feature vectors of the newly added playlists in the music library to obtain corresponding similarity values; the playlist feature vector is a feature representation of the deep semantic information of the graphic and text information of the corresponding newly added playlist; the playlist distribution module is used to distribute multiple newly added playlists whose similarity values meet the preset conditions to the user, and correspondingly obtain the user behavior data corresponding to the access of the multiple newly added playlists by the user; the playlist screening module is used to screen out target newly added playlists according to the access popularity of each newly added playlist statistically obtained from the user behavior data corresponding to the access of all newly added playlists in the music library by different users, and construct the target newly added playlists into a recommendation list.

[0032] In a further embodiment, the user processing module includes:

[0033] A data acquisition sub-module, which is used to obtain the mapping relationship data between the unique feature information of the playlists and songs accessed by the user and the behavior types implemented by the user from the historical behavior data of the user. The behavior types include any one or more of click, play, favorite, download, and share;

[0034] A data vectorization sub-module, which is used to encode according to the mapping relationship data to obtain a user encoding vector;

[0035] A feature extraction sub-module, which is used to adopt a pre-trained and converged user feature extraction model to extract features according to the user encoding vector to obtain a corresponding user feature vector.

[0036] In a further embodiment, the playlist distribution module includes:

[0037] A threshold detection sub-module, which is used to detect whether the total historical distribution amount corresponding to the multiple newly added playlists reaches a preset threshold. When it reaches, stop distributing the corresponding newly added playlists, otherwise distribute the corresponding newly added playlists to the user;

[0038] A user monitoring sub-module, which is used to monitor the user behavior data implemented by the user based on the distributed newly added playlists;

[0039] A data acquisition sub-module, which is used to detect and obtain the user behavior data for the distributed newly added playlists in response to the user behavior data submitted by the user.

[0040] In a further embodiment, the playlist screening module includes:

[0041] A visit volume statistics sub-module, which is used to, for all newly added playlists in the music library, according to the user behavior data generated by being accessed by different users, statistically calculate the visit volume corresponding to different behavior types imposed on each newly added playlist, and the behavior types include any one or more of click, play, favorite, download, and share;

[0042] A visit heat calculation sub-module, which is used to, for each newly added playlist, perform a weighted sum of the visit volumes of different behavior types to obtain the corresponding visit heat of each newly added playlist;

[0043] A threshold judgment sub-module, which is used to judge whether the visit heat of each newly added playlist exceeds a preset threshold, and determine the newly added playlists with visit heat exceeding the preset threshold as target newly added playlists;

[0044] A playlist recommendation sub-module, which is used to create a recommendation list and add the target newly added playlists to the recommendation list.

[0045] In a further embodiment, after the playlist screening module, there is also included:

[0046] A playlist push sub-module, which is used to respond to a user's playlist recommendation request, and according to the user's user feature vector, match one or more preferred newly added playlists from the newly added playlists in the recommendation list whose playlist feature vectors are similar to the user feature vector, and push the preferred newly added playlists to the corresponding users.

[0047] In a further embodiment, before the user processing module, there is also included:

[0048] A model training sub-module, which is used to train a dual tower model for extracting the user feature vector and the playlist feature vector by using a predicted data set to make it reach a convergence state; wherein, the data set includes a plurality of data samples, each data sample represents the mapping relationship data between the unique feature information of the playlists and songs accessed by a corresponding user and the behavior types implemented by the user, and the playlist feature data is the graphic and text information of the playlists accessed by the corresponding user;

[0049] A model invocation sub-module, which is used to invoke the dual tower model that has been trained to a convergence state, and extract the corresponding playlist feature vector according to the graphic and text information of any playlist in the music library.

[0050] A computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory. The central processing unit is used to call and run a computer program stored in the memory to execute the steps of the new playlist recommendation method described in the present application.

[0051] A computer-readable storage medium provided to meet another purpose of the present application stores a computer program implemented based on the new playlist recommendation method in the form of computer-readable instructions. When the computer program is called and run by a computer, it executes the steps included in the method.

[0052] A computer program product provided to meet another purpose of the present application includes a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the steps of the method described in any embodiment of the present application.

[0053] Compared with the prior art, the advantages of the present application are as follows: When each user needs the platform to recommend playlists, for a large number of new playlists in the platform, according to the similarity between the user feature vector representing user behavior data and the playlist feature vectors corresponding to these new playlists, distribute playlists with higher similarity to the user, so that new playlists can be distributed to potentially matching users. Then, the platform uses a listening mechanism at the back end to control the total distribution volume of each new playlist, and correspondingly obtain the user behavior data submitted by the users who receive the new playlist. Through these user behavior data, count the access situations of each new playlist, quantitatively represent it as the access popularity of the new playlist. Finally, determine the pros and cons of each new playlist according to the access popularity, make a selection, construct a recommendation list, and utilize the voting effect of user interaction data to realize the evaluation and optimization of new playlists, eliminating the trouble of manual playlist screening and greatly improving the playlist output quality of the online music platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The above and / or additional aspects and advantages of the present application will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0055] Figure 1 is a schematic flowchart of a typical embodiment of the new playlist recommendation method of the present application;

[0056] Figure 2 is a schematic flowchart of the working process of the neural network model on the user side in the dual tower model in the embodiment of the present application;

[0057] Figure 3 is a schematic flowchart of the distribution of new playlists in the embodiment of the present application;

[0058] Figure 4 is a schematic flowchart of creating a recommendation list in the embodiment of the present application;

[0059] Figure 5 It is a schematic flowchart for adding post-playlist recommendations in the embodiments of this application;

[0060] Figure 6 It is a schematic flowchart for adding playlist feature vectors of pre-extracted playlists in the embodiments of this application;

[0061] Figure 7 It is a principle block diagram of the newly added playlist recommendation device of this application;

[0062] Figure 8 It is a schematic structural diagram of a computer device adopted by this application. Detailed implementation manners

[0063] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as a limitation to the present application.

[0064] Those skilled in the art of this technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the", and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the specification of the present application means the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any unit and all combinations of one or more related listed items.

[0065] Those skilled in the art of this technology can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.

[0066] Those skilled in the art of the present technology can understand that the "client", "terminal", and "terminal device" used herein include both devices with wireless signal receivers that only have the ability to receive and no ability to transmit, and devices with receiving and transmitting hardware that have the receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, etc., which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, pager, Internet / intranet access, web browser, notepad, calendar, and / or GPS (Global Positioning System) receiver; conventional laptop and / or palm-top computers or other devices, which are conventional laptop and / or palm-top computers or other devices with and / or including a radio frequency receiver. The "client", "terminal", and "terminal device" used herein can be portable, transportable, installed in a vehicle (air, sea, and / or land), or suitable for and / or configured to run locally, and / or run in a distributed manner at any other location on the earth and / or in space. The "client", "terminal", and "terminal device" used herein can also be a communication terminal, an Internet access terminal, a music / video playback terminal, for example, it can be a PDA, MID (Mobile Internet Device), and / or a mobile phone with music / video playback function, or can also be a smart TV, a set-top box, and other devices.

[0067] The hardware referred to by names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer, and is a hardware device with the necessary components disclosed by the von Neumann principle, including a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit loads the program stored in the external memory into the memory for execution, executes the instructions in the program, and interacts with the input / output devices to complete specific functions.

[0068] It should be noted that the concept of "server" in this application can similarly be extended to apply to server clusters. According to the network deployment principles understood by those skilled in the art, the various servers should be logically divided. Physically, these servers can either be independent of each other but can be called through interfaces, or can be integrated into a single physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility and should not be restricted by this when implementing the network deployment method of this application.

[0069] One or several technical features of this application, unless expressly specified, can either be deployed on the server and accessed by the client remotely invoking the online service interface provided by the server, or can be directly deployed and run on the client for access.

[0070] The neural network models cited or possibly cited in this application, unless expressly specified, can either be deployed on a remote server and remotely invoked by the client, or can be directly invoked on the client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0071] All kinds of data involved in this application, unless expressly specified, can either be remotely stored on the server or stored on the local terminal device, as long as it is suitable for being invoked by the technical solution of this application.

[0072] Those skilled in the art should be aware of this: Although the various methods of this application are described based on the same concept and thus show commonality with each other, unless otherwise specified, these methods can all be executed independently. Similarly, for the various embodiments disclosed in this application, they are all proposed based on the same inventive concept. Therefore, for concepts with the same expression, as well as concepts that are only appropriately transformed for convenience although the concept expressions are different, they should be equivalently understood.

[0073] For the various embodiments to be disclosed in this application, unless expressly stated to be mutually exclusive, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct new embodiments, as long as this combination does not deviate from the creative spirit of this application and can meet the requirements in the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0074] A new playlist recommendation method of this application can be programmed into a computer program product and deployed to run on the server. Thereby, the client can access the interface opened after the computer program product runs in the form of a web program or an application program, and achieve human-computer interaction with the process of the computer program product through the graphical user interface.

[0075] Please refer to Figure 1 , in a typical embodiment of the new playlist recommendation method of the present application, the following steps are included:

[0076] Step S1100: Respond to the user's recommendation request, and obtain the user feature vector according to the historical behavior data of the user. The user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data;

[0077] The historical behavior data of the user are operation behaviors such as clicking, playing, favoriting, downloading, sharing, etc. implemented on the graphical user interface of the playlists or songs accessed by the user in a recent period of time. Further, the operation behaviors reflect the user's preference for the playlists or songs to a certain extent, and also include the user's preference for any one or more characteristic attributes among the era genre, singer, lyrics, language, etc. corresponding to the playlists or songs. Relatively speaking, the historical behavior data implies the user's preference for accessing playlists and songs in the music library.

[0078] The user clicks on the control of the relevant recommended playlist on the graphical user interface of the application installed on the terminal device, such as scrolling the graphical user interface to the recommended area, clicking on the more button control in the recommended playlist column displayed on the home page interface, clicking on the play mode switching control on the graphical user interface of the selected song to set the play mode to the heart mode, etc. Thus, a request for recommending playlists or songs for the user is triggered. This request is sent to the server of the present application, and after receiving the request, the server responds to it.

[0079] The server responds to the request, and according to the mapping association relationship between the historical behavior data of the user and the pre-collected user feature vector, obtains the user feature vector from the cache. The pre-collected user feature vector is obtained by calling the text feature extraction model on the user side in the twin tower model that has been pre-trained to the convergence state, performing feature representation on the encoded vector formed by encoding the historical behavior data, and obtaining the feature vector corresponding to the deep semantic information therein. Then it is stored in the cache database to facilitate the quick matching process between the user side and the feature vector corresponding to the new playlist side in the cache, quickly calling the user feature vector for similarity calculation, and finally realizing the matching of the corresponding new playlist according to the user's historical behavior data. The text feature extraction model can be implemented by using various pre-trained neural network models based on CNN and RNN in the prior art.

[0080] The dual - tower model consists of a text feature extraction model on the user side and a neural network model on the new song list side. This model pre - takes the user's behavior data and the corresponding song and playlist data accessed by the user as the training set and sets the test set accordingly, and thus, conducts supervised training until it converges. More technical details about the dual - tower model are revealed in a subsequent embodiment of this application, which will not be elaborated here for the time being.

[0081] In other embodiments, other neural network models can also be used to represent the user feature vector, which should be known to those skilled in the art.

[0082] Step S1200: Calculate the similarity between the user feature vector and the pre - collected playlist feature vectors of the new playlists in the music library to obtain the corresponding similarity values; the playlist feature vector is the feature representation of the deep semantic information of the graphic and text information of the corresponding new playlist.

[0083] The new playlist is a playlist composed of one or more songs in an online music platform. The playlist can be created by platform users or professionals and submitted to the music platform in ways such as online upload, email sending, on - site submission, etc. Then, it is collected by the relevant responsible personnel in the music platform for certain pre - processing and finally uniformly converted into digital music form for uploading and entered into the music library of the music platform.

[0084] The pre - collected playlist feature vector is to call the neural network model on the song list side in the dual - tower model that has been pre - trained to the convergence state, perform feature representation on the encoded vector formed by encoding the graphic and text information of the new playlist in the music library, obtain the feature vector corresponding to the deep semantic information, and then store it in the cache database. The neural network model can be implemented by using a variety of pre - trained neural network models based on CNN and RNN in the prior art.

[0085] Call the preset similarity calculation method to calculate the similarity between the user feature vector and each of the corresponding playlist feature vectors in the music library, and thus, obtain the similarity values between the feature vectors corresponding to the user side and the new song list side in the cache.

[0086] Step S1300: Distribute multiple new playlists whose similarity values meet the preset conditions to the user, and correspondingly obtain the user behavior data corresponding to the access of the user to the multiple new playlists.

[0087] The preset condition is a preset similarity threshold. Specifically, the similarity threshold can be an empirical threshold or an experimental threshold, which can be selected and set by those skilled in the art as needed.

[0088] Filter out the newly added playlists corresponding to the playlist feature vectors whose similarity values ​​exceed a preset similarity threshold, and push the multiple newly added playlists to the user. The push may be to invite the user to participate in a free trial of new songs, display the newly added playlist on the activity interface, or simply recommend a playlist in the usual way, etc. Furthermore, set corresponding monitoring on the newly added playlist display interface to collect user behavior data corresponding to the user's access to the newly added playlist.

[0089] Step S1400, filter out target new playlists based on the visit popularity of each new playlist obtained by counting the corresponding user behavior data of all new playlists in the music library being visited by different users, and construct the target new playlists into a recommendation list.

[0090] The aforementioned process from step S1100 to step S1300 can respond to any user of the platform, thereby enabling the platform to distribute new playlists to a large number of platform users. Correspondingly, the platform server can be responsible for monitoring, counting and controlling the number of each new playlist distributed. For example, for each new playlist, limited distribution can be performed, and the distribution of the corresponding new playlist can be stopped when the total distribution reaches a preset value.

[0091] The platform server monitors and collects the behavioral data of each platform user's access to the corresponding newly added playlists, thereby counting the access popularity of each newly added playlist. Furthermore, the corresponding newly added playlists are sorted from high to low according to the access popularity, and the newly added playlists with higher rankings are selected as the target newly added playlists. The corresponding rankings of each target playlist are constructed into a recommendation list.

[0092] Through the disclosure of this typical embodiment, it can be seen that the present application has many technical advantages, including but not limited to: when each user needs the platform to recommend a playlist, for a large number of newly added playlists in the platform, based on the similarity between the user feature vector representing the user behavior data and the playlist feature vectors corresponding to these newly added playlists, playlists with higher similarity are distributed to users, so that the newly added playlists can be distributed to potential matching users. Then, the platform uses a monitoring mechanism at the back end to control the total distribution of each newly added playlist, and correspondingly obtains the user behavior data subsequently submitted by the user who receives the newly added playlist. Through these user behavior data, the access situation of each newly added playlist is counted, and quantitatively expressed as the access popularity of the newly added playlist. Finally, the pros and cons of each newly added playlist are determined according to the access popularity, and a screening is made to construct a recommendation list. The voting effect of user interaction data is used to evaluate and select the newly added playlists, eliminating the trouble of manual screening of playlists and greatly improving the quality of playlist output of the online music platform.

[0093] See also Figure 2, in a further embodiment, obtaining the user feature vector according to the historical behavior data of the user includes the following steps:

[0094] Step S1110: Obtain the mapping relationship data between the uniqueness feature information of the playlists and songs accessed by the user and the behavior types implemented by the user from the historical behavior data of the user. The behavior types include any one or more of click, play, favorite, download, and share;

[0095] The uniqueness feature information of the playlists and songs is the uniqueness feature information created correspondingly as an identifier according to the corresponding feature information generation rules of the platform when the playlists and songs are first uploaded by the user or the platform. The identifiers corresponding to different playlists and songs are different, which is convenient for software engineering technicians to store and call the playlists and songs.

[0096] The historical behavior data of the user is the behavior type data implemented when the user accesses the playlists and songs. The behavior type data is collected by a listener preset in the corresponding display interface of the playlists and songs and uploaded to the server. Then, the server maps and associates the behavior type data with the uniqueness feature information of the playlists and songs as behavior data and stores it in the behavior database corresponding to the user.

[0097] Obtain the uniqueness feature information from the historical behavior data of the user, and thus obtain the corresponding mapping relationship data, that is, the behavior type data implemented by the user.

[0098] In a typical embodiment, different unique feature information lists of playlists and songs can be constructed using different behavior types. Each playlist and song list represents the playlists and songs of interest to the user corresponding to its respective behavior type, and subsequent encoding can be performed according to these lists.

[0099] Step S1120: Obtain the user encoding vector by encoding according to the mapping relationship data;

[0100] Call the text feature extraction model on the user side in the twin tower model pre-trained to the convergence state, and perform one-hot encoding operation on the mapping relationship data, that is, the behavior type data, specifically, for example, the playlist and song lists corresponding to various behavior types, to obtain the user encoding vector. The text feature extraction model is preferably a mature model such as Bert or Electra, and those skilled in the art can flexibly adopt it.

[0101] Step S1130: Adopt a user feature extraction model pre-trained to convergence, and perform feature extraction according to the user encoding vector to obtain the corresponding user feature vector.

[0102] Further, the text feature extraction model on the user side in the two-tower model pre-trained to the convergence state is called to extract features from the user encoding vector. Correspondingly, direct normalization operations are performed on the dense features, and the sparse features are embedded and dimension-reduced to a low-dimensional space. Then, the dense and sparse features are concatenated to obtain the corresponding user feature vector, and the user feature vector is associated with the user and stored in the in-memory database of the cache.

[0103] In this embodiment, the user feature vector is constructed according to the user's behavior data and pre-stored in the cache. Therefore, when the user feature vector needs to be called, it can be quickly obtained from the cache, which is extremely suitable for the subsequent online fast recall business scenario for matching. By obtaining the playlist and song list that the user is interested in according to the user's historical behavior data, the personal preferences of the user can be effectively characterized, which is convenient for recommending new playlists that match the interests for the user subsequently.

[0104] Please refer to Figure 3 , in a further embodiment, a plurality of new playlists whose similarity values meet the preset conditions are distributed to the user, and the user behavior data corresponding to the access of the plurality of new playlists by the user is obtained accordingly, including the following steps:

[0105] Step S1310: Detect whether the total historical distribution volume corresponding to the plurality of new playlists reaches a preset threshold. When it reaches, stop distributing the corresponding new playlists; otherwise, distribute the corresponding new playlists to the user;

[0106] The total historical distribution volume is the number of users distributed for a new playlist. Accordingly, the preset threshold can be set. The simplest way is to set an empirical fixed value, but it can also be set more complexly based on the principle of effective statistics in statistics. For example: taking 20% of the total number of daily active users statistically on the platform as the threshold; screening out a batch of high-quality playlists with the fastest soaring like counts in a recent period on the platform, statistically calculating their respective like counts and calculating the corresponding average value, and further multiplying the average value by 1.25 times as the threshold. Here, those skilled in the art can set it flexibly as needed.

[0107] The server responds to the new playlist distribution event, statistically calculates the number of distributed users corresponding to the new playlist in real time, and compares the number of users with the preset threshold. Thus, the control of the new playlist distribution is realized, and then all new playlists in the music library are distributed to the corresponding users to complete the new playlist distribution event.

[0108] Step S1320: Monitor the user behavior data implemented by the user for the distributed new playlists;

[0109] In one embodiment, after the server distributes a new playlist to a user, it starts a monitoring mechanism for that user. When the user performs an operation on the received new playlist, it triggers the submission of corresponding user behavior data to the server, and the server can collect the corresponding behavior data of the user. These user behavior data can correspond to events such as liking, favoriting, downloading, and sharing on the new playlist playing interface. Thus, when the user triggers a corresponding event and the server determines that the corresponding new playlist has been liked, favorited, downloaded successfully, or shared successfully, it can collect the corresponding behavior data of the user.

[0110] Step S1330: In response to the user behavior data submitted by the user, detect and obtain the user behavior data for the distributed new playlist.

[0111] In essence, the server can access the user behavior data generated and submitted by any user at any time. However, for statistical purposes, the server can detect the user behavior data associated with the distributed new playlist and the unique identifier of the corresponding user, and then store it in the cache for subsequent statistics.

[0112] In this embodiment, since the new playlist has not been manually reviewed and rated as a high-quality playlist by relevant professionals on the platform, but instead the threshold is reasonably set to control the distribution of new playlists. The twin-tower model of this application selects a suitable new playlist based on the user's historical behavior data and then performs corresponding push. That is, for the new playlist distributed to this user, this user is suitable to be the reviewer of this new playlist and can "vote" on this new playlist. The new playlist server is responsible for collecting and statistically analyzing the user behavior data triggered and submitted by users for these new playlists. Relatively speaking, it can collect the feedback of the public review for each new playlist. Generally speaking, on the one hand, given the certain uncertainty in the quality of new playlists, directly distributing new playlists to a large number of corresponding users without restriction may cause some doubts among users about the unstable quality of the songs pushed by the platform and thus reduce their favorability. Therefore, the method of reasonably setting the threshold is beneficial to both ensuring a certain number of users to review the corresponding new playlists and reducing the adverse impact. On the other hand, handing it over to the public review is more objective, has a better review effect, and saves labor costs compared to manual review by relevant professionals on the platform. Therefore, the high-quality new playlists evaluated in this way are more suitable for the majority of users.

[0113] Please refer to Figure 4 In a further embodiment, step S1400: Screen out the target new playlists according to the access popularity of each new playlist statistically obtained from the user behavior data corresponding to the full amount of new playlists in the music library being accessed by different users, and construct the target new playlists into a recommended list, including the following steps:

[0114] Step S1410: For all newly added playlists in the music library, according to the user behavior data generated by being accessed by different users, count the access volume corresponding to different behavior types imposed on each newly added playlist. The behavior types include any one or more of click, play, favorite, download, and share.

[0115] Call the user behavior data collected in the cache when listening to the access of all newly added playlists by different users. Accordingly, count the number of times each newly added playlist is clicked, the number of times it is played (including being repeatedly played by the same user and being played by different users, the number of likes, the number of favorites, the number of successful downloads, and the number of successful shares. These numbers are the so-called access volume.

[0116] Step S1420: For each newly added playlist, perform a weighted sum of the access volumes of different behavior types to obtain the access popularity corresponding to each newly added playlist.

[0117] Based on the access volumes corresponding to different behavior types, namely the number of clicks, the number of plays, the number of likes, the number of favorites, the number of successful downloads, and the number of successful shares, and different weights corresponding to different behavior types, perform a corresponding weighted sum as the access popularity corresponding to each newly added playlist.

[0118] The weights are: 1 point for click, 2 points for play, 3 points for like, 5 points for favorite, 6 points for download, and 9 points for share. Those skilled in the art can flexibly set the weights according to the input degree of the behavior types, such as the input time cost, social pressure, and implementation complexity.

[0119] Step S1430: Determine whether the access popularity of each newly added playlist exceeds a preset threshold, and determine the newly added playlists with access popularity exceeding the preset threshold as target newly added playlists.

[0120] In one embodiment, the preset threshold is to count the number of clicks, the number of plays, the number of likes, the number of favorites, the number of successful downloads, and the number of successful shares corresponding to a batch of high-quality playlists in the recent period, perform a corresponding weighted sum calculation according to the weights, and then calculate the average value of the calculation result as the preset threshold. The setting of the threshold here can be flexibly changed by those skilled in the art as needed.

[0121] After determining the preset threshold, it can be compared with the access popularity to select the best among each newly added playlist, and determine the newly added playlists exceeding the preset threshold as target newly added playlists.

[0122] In an alternative implementation, each newly added playlist may be classified according to the content of the playlist and the songs, and then for each category, a newly added playlist with the highest access popularity may be selected, which may be specifically selected by a preset threshold value as the target newly added playlist.

[0123] Step S1440: create a recommendation list, and add the target new playlist to the recommendation list.

[0124] Further, the target newly added song list is sorted in descending order of popularity according to the access popularity of the target newly added song list, and then the sorted target newly added song list is added to the recommendation list.

[0125] In this embodiment, the user behavior data is further quantified as the access popularity of the corresponding newly added playlist, which is then compared with the access popularity corresponding to the current high-quality playlist. Thus, the target newly added playlist with higher access popularity is reasonably selected and the recommendation list is constructed accordingly. At this point, the newly added playlist is selected based on the behavioral data of public review.

[0126] See also Figure 5 In a further embodiment, after the step of constructing the target newly added song list as a recommendation list, the following steps are included:

[0127] Step S1500, respond to the user's request for a playlist recommendation, match one or more preferred new playlists whose playlist feature vectors are similar to the user feature vector from the new playlists in the recommendation list according to the user feature vector of the user, and push the preferred new playlists to the corresponding user.

[0128] The user clicks on the relevant recommended playlist control on the graphical user interface of the application installed on his terminal device, triggering a request for recommending a playlist or songs for the user. The request is sent to the server of this application, and the server responds to the request after receiving it.

[0129] The server responds to the request and obtains the user feature vector of the user in the cache according to the unique identification code of the user, and then calculates the similarity between the user feature vector and the playlist feature vector corresponding to the newly added playlist in the recommendation list in the cache, and screens out the playlist feature vectors whose similarity exceeds a preset threshold. The preset threshold here can be flexibly set by technical personnel in this field according to the actual recall results, and further determines that the corresponding newly added playlist is the preferred newly added playlist, and then pushes the preferred newly added playlist to the user.

[0130] In this embodiment, through the neural network models on the user side and the new playlist side in the two-tower model pre-trained to the convergence state, the user feature vectors and playlist feature vectors are pre-collected accordingly and stored in the in-memory database in the cache. Furthermore, during the matching process between the user side and the new playlist side, the required feature vectors can be quickly called from the cache for similarity calculation to achieve efficient matching.

[0131] Please refer to Figure 6 , in a further embodiment, before responding to a user recommendation request, the following steps are included:

[0132] Step S1000: Use a predicted data set to train a two-tower model for extracting the user feature vectors and playlist feature vectors until it reaches the convergence state; wherein, the data set includes multiple data samples, and each data sample represents the mapping relationship data between the user feature data corresponding to a user and the playlist feature data. The user feature data is the mapping relationship data between the unique feature information of the playlists and songs accessed by the corresponding user and the behavior types implemented by the user, and the playlist feature data is the graphic and text information of the playlists accessed by the corresponding user.

[0133] The cover image features of the playlist can be extracted by a neural network model based on CNN, and the corresponding image feature vectors can be constructed therefrom; the text feature models of neural network models based on RNN, such as Bert, Electra, etc., can be used to extract the features corresponding to the text information such as the playlist title, playlist introduction, singer name, album name, etc. of the playlist, and the corresponding text feature vectors can be constructed therefrom. The dense features among the features are subjected to one-hot encoding operations, and the sparse features are dimension-reduced to a low-dimensional space by embedding, and then feature splicing operations are performed. Further, the image feature vectors and the text feature vectors are spliced to construct a graphic and text feature vector as the playlist feature vector of the playlist.

[0134] The model supervised training of the dual - tower model is carried out separately on both sides, namely the user side and the new song side. Two corresponding neural network models are used respectively. At the input layer stage of the model, the neural network model on the user side is responsible for constructing the user feature vector, while the neural network model on the new song side is responsible for constructing the playlist feature vector. At the representation layer stage, the user feature vector and the playlist feature vector are respectively subjected to further feature extraction through two fully - connected layers in their respective corresponding neural network models and transformed into fixed - length vectors, thereby obtaining a low - dimensional user feature vector and a low - dimensional playlist feature vector with the same dimension. After the model training converges, they are stored in an in - memory database such as Redis, that is, in the cache. The number of network layers and dimensions in their respective corresponding neural network models can be different, but the dimensions of the respectively output feature vectors must be the same in order to perform operations in the matching layer. The dimensions of the low - dimensional user feature vector and the low - dimensional playlist feature vector can be 32 - dimensional or 128 - dimensional, etc. At the matching layer stage of the model, the cosine similarity is calculated by calling the low - dimensional user feature vector and the low - dimensional playlist feature vector. The result obtained by the cos function is input into the sigmoid function to calculate the logloss with the true label corresponding to the test set, and check whether the dual - tower model converges. The auc index is mainly used for evaluation. Further adjust the relevant parameters until the dual - tower model converges to complete the supervised training. The test set is flexibly set by those skilled in the art according to the actual scenario.

[0135] Step S1010: Call the dual - tower model that has been trained to the convergence state, and for any playlist in the music library, extract the corresponding playlist feature vector according to the graphic and text information of the playlist.

[0136] The cover image features of the playlist can be extracted by a neural network model based on CNN, thereby constructing a corresponding image feature vector. The text feature models of RNN - based neural network models such as Bert and Electra can be used to extract the features corresponding to the playlist title, playlist introduction, singer name, album name, etc. of the playlist, thereby constructing corresponding text feature vectors. The dense features in the features are subjected to one - hot encoding operations, and the sparse features are reduced to a low - dimensional space by embedding, and then feature splicing operations are performed. Further, the image feature vector and the text feature vector are spliced to construct a graphic - text feature vector and input into the input layer of the dual - tower model, and then the playlist feature vector corresponding to the playlist is obtained.

[0137] In this embodiment, the dual - tower model is trained in a supervised manner, so that the accuracy of the similarity result output by it is relatively high, and the matching accuracy is also higher.

[0138] Please refer to Figure 7, a new playlist recommendation device provided by this application is functionally deployed in accordance with the new playlist recommendation method of this application, including: a user processing module 1100, a similarity matching module 1200, a playlist distribution module 1300, and a playlist screening module 1400. Among them, the user processing module is used to respond to a user recommendation request and obtain the user feature vector according to the historical behavior data of the user. The user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data; the similarity matching module is used to calculate the similarity between the user feature vector and the pre-collected playlist feature vectors of the new playlists in the music library to obtain the corresponding similarity value; the playlist feature vector is the feature representation of the deep semantic information of the graphic and text information of the corresponding new playlist; the playlist distribution module is used to distribute multiple new playlists whose similarity values meet the preset conditions to the user, and correspondingly obtain the user behavior data corresponding to the access of the multiple new playlists by the user; the playlist screening module is used to screen out the target new playlists according to the access popularity of each new playlist statistically obtained from the user behavior data corresponding to the access of all new playlists in the music library by different users, and construct the target new playlists into a recommendation list.

[0139] In a further embodiment, the user processing module 1100 includes: a data acquisition sub-module, which is used to obtain the mapping relationship data between the unique feature information of the playlists and songs accessed by the user and the behavior types implemented by the user from the historical behavior data of the user, and the behavior types include any one or more of click, play, favorite, download, and share; a data vectorization sub-module, which is used to encode according to the mapping relationship data to obtain a user encoding vector; a feature extraction sub-module, which is used to adopt a pre-trained user feature extraction model that has converged to extract features according to the user encoding vector to obtain the corresponding user feature vector.

[0140] In a further embodiment, the playlist distribution module 1300 includes: a threshold detection sub-module, which is used to detect whether the total historical distribution of the multiple new playlists reaches a preset threshold. When it reaches, stop distributing the corresponding new playlists, otherwise distribute the corresponding new playlists to the user; a user monitoring sub-module, which is used to monitor the user behavior data implemented by the user based on the distributed new playlists; a data acquisition sub-module, which is used to detect and obtain the user behavior data for the distributed new playlists in response to the user behavior data submitted by the user.

[0141] In a further embodiment, the playlist screening module 1400 includes: a visit volume statistics sub-module, which is used to, for all newly added playlists in the music library, according to the user behavior data generated by being accessed by different users, count the visit volume corresponding to different behavior types imposed on each newly added playlist, and the behavior types include any one or more of click, play, favorite, download, and share; a visit popularity calculation sub-module, which is used to, for each newly added playlist, perform weighted summation on the visit volumes of different behavior types to obtain the visit popularity corresponding to each newly added playlist; a threshold judgment sub-module, which is used to judge whether the visit popularity of each newly added playlist exceeds a preset threshold, and determine the newly added playlists with visit popularity exceeding the preset threshold as target newly added playlists; a playlist recommendation sub-module, which is used to create a recommendation list and add the target newly added playlists to the recommendation list.

[0142] In a further embodiment, after the playlist screening module 1100, there is further included: a playlist push sub-module, which is used to respond to a user's playlist recommendation request, and according to the user feature vector of the user, match one or more preferred newly added playlists whose playlist feature vectors are similar to the user feature vector from the newly added playlists in the recommendation list, and push the preferred newly added playlists to the corresponding users.

[0143] In a further embodiment, before the user processing module 1400, there is further included: a model training sub-module, which is used to train a dual tower model for extracting the user feature vector and the playlist feature vector by using a predicted data set until it reaches a converged state; wherein, the data set includes multiple data samples, and each data sample represents the mapping relationship data between the unique feature information of the playlists and songs accessed by a corresponding user and the behavior types implemented by the user, and the playlist feature data is the graphic and text information of the playlists accessed by the corresponding user; a model invocation sub-module, which is used to invoke the dual tower model that has been trained to a converged state, and extract the corresponding playlist feature vector according to the graphic and text information of any playlist in the music library.

[0144] To solve the above technical problems, an embodiment of the present application further provides a computer device. As Figure 8As shown in the figure, it is a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected through a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database can store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a method for recommending new playlists. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device can store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the method for recommending new playlists of this application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 8 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0145] In this embodiment, the processor is used to execute Figure 8 the specific functions of each module and its sub-modules in the figure. The memory stores the program code and various types of data required to execute the above modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. The memory in this embodiment stores the program code and data required to execute all modules / sub-modules in the new playlist recommendation device of this application. The server can call the program code and data of the server to execute the functions of all sub-modules.

[0146] This application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to execute the steps of the method for recommending new playlists according to any embodiment of this application.

[0147] This application also provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by one or more processors, the steps of the method according to any embodiment of this application are implemented.

[0148] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. This computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0149] In summary, the present application matches corresponding newly added playlists according to the user feature vectors characterizing user behavior data, and obtains the corresponding access popularity by "rating" the newly added playlists by the user. Thus, high-quality newly added playlists with higher access popularity are screened out to construct a recommendation list, making use of the voting effect of a sufficient amount of user interaction data to objectively and efficiently select high-quality newly added playlists, saving a large amount of labor costs, and greatly improving the playlist production efficiency and quality of the online music platform.

[0150] Those skilled in the art of the present technology can understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, the other steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and solutions in the prior art that are the same as those disclosed in the various operations, methods, and processes in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0151] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art of the present technology, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for recommending new playlists, characterized in that, it includes the following steps: Respond to a user's recommendation request, and obtain the user feature vector according to the user's historical behavior data. This user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data; Calculate the similarity between the user feature vector and the pre-collected playlist feature vectors of the new playlists in the music library to obtain the corresponding similarity value; this playlist feature vector is the feature representation of the deep semantic information of the corresponding graphic and text information of the new playlist; Distribute multiple new playlists whose similarity values meet the preset conditions to this user, and correspondingly obtain the user behavior data corresponding to the access of these multiple new playlists by this user, including: detecting whether the total historical distribution volume corresponding to these multiple new playlists reaches the preset threshold, and stopping distributing the corresponding new playlists when it reaches, otherwise distributing the corresponding new playlists to this user; monitoring the user behavior data implemented by this user based on the distributed new playlists; in response to the user behavior data submitted by this user, detecting and obtaining the user behavior data for the distributed new playlists; Screen out the target new playlists according to the access popularity of each new playlist statistically obtained from the user behavior data corresponding to the access of all new playlists in the music library by different users, and construct the target new playlists into a recommendation list; Respond to the user's playlist recommendation request, and according to the user feature vector of this user, match one or more preferred new playlists whose playlist feature vectors are similar to the user feature vector from the new playlists in the recommendation list, and push the preferred new playlists to the corresponding user.

2. The method for recommending new playlists according to claim 1, characterized in that, obtaining the user feature vector according to the user's historical behavior data includes the following steps: Obtain the mapping relationship data between the unique feature information of the playlists and songs accessed by this user and the behavior types implemented by this user from the user's historical behavior data, and the behavior types include any one or more of click, play, favorite, download, and share; Encode according to the mapping relationship data to obtain a user encoding vector; Adopt a pre-trained user feature extraction model to extract features according to the user encoding vector to obtain the corresponding user feature vector.

3. The method for recommending new playlists according to claim 1, characterized in that, screening out the target new playlists according to the access popularity of each new playlist statistically obtained from the user behavior data corresponding to the access of all new playlists in the music library by different users, and constructing the target new playlists into a recommendation list includes the following steps: For all new playlists in the music library, according to the user behavior data generated by being accessed by different users, count the access volume corresponding to each new playlist for different behavior types, and the behavior types include any one or more of click, play, favorite, download, and share; For each new playlist, perform a weighted sum of the access volumes of different behavior types to obtain the access popularity corresponding to each new playlist; Determine whether the access popularity of each newly added playlist exceeds a preset threshold, and determine the newly added playlists with access popularity exceeding the preset threshold as target newly added playlists; Create a recommendation list and add the target newly added playlists to the recommendation list.

4. The method for recommending newly added playlists according to claim 1, characterized in that, before responding to a user recommendation request, the following steps are included: Using a predicted data set to train a dual tower model for extracting the user feature vector and the playlist feature vector until it reaches a converged state; wherein, the data set includes a plurality of data samples, each data sample represents the mapping relationship data between the unique feature information of the playlists and songs accessed by a corresponding user and the behavior type implemented by the user, and the playlist feature data is the graphic and text information of the playlists accessed by the corresponding user; Call the dual tower model that has been trained to a converged state, and extract the corresponding playlist feature vector for any playlist in the music library according to the graphic and text information of the playlist.

5. A device for recommending newly added playlists, characterized in that, comprising: A user processing module, configured to respond to a user recommendation request, and obtain a user feature vector according to the historical behavior data of the user, where the user feature vector is used to represent the deep semantic information corresponding to the user's preference for accessing playlists and songs in the music library implied by the historical behavior data; A similarity matching module, configured to calculate the similarity between the user feature vector and the pre-collected playlist feature vectors of newly added playlists in the music library to obtain corresponding similarity values; the playlist feature vector is a feature representation of the deep semantic information of the graphic and text information of the corresponding newly added playlist; A playlist distribution module, configured to distribute multiple newly added playlists whose similarity values meet preset conditions to the user, and correspondingly obtain the user behavior data corresponding to the access of the multiple newly added playlists by the user, including: detecting whether the total historical distribution volume corresponding to the multiple newly added playlists reaches a preset threshold, and stopping distributing the corresponding newly added playlists when it reaches, otherwise distributing the corresponding newly added playlists to the user; monitoring the user behavior data implemented by the user based on the distributed newly added playlists; in response to the user behavior data submitted by the user, detecting and obtaining the user behavior data for the distributed newly added playlists; A playlist screening module, configured to screen out target newly added playlists according to the access popularity levels of each newly added playlist statistically obtained from the user behavior data corresponding to the access of all newly added playlists in the music library by different users, and construct the target newly added playlists into a recommendation list; A playlist pushing sub-module, configured to respond to a user's playlist recommendation request, and match one or more preferred newly added playlists whose playlist feature vectors are similar to the user feature vector from the newly added playlists in the recommendation list, and push the preferred newly added playlists to the corresponding user.

6. A computer device, including a central processing unit and a memory, characterized in that, The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, it stores a computer program implemented according to the method according to any one of claims 1 to 4 in the form of computer-readable instructions, and when the computer program is called and run by a computer, it executes the steps included in the corresponding method.

8. A computer program product, comprising a computer program / instructions, characterized in that, when the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images