CN113836401B - Playlist recommendation method, device and readable storage medium - Google Patents

Abstract

The invention discloses a song list recommending method, a device and a readable storage medium, wherein the song list recommending method comprises the following steps: acquiring a history song list of a user, and respectively acquiring characteristic data of the history song list, characteristic data of each history song contained in the history song list and characteristic data of the user; determining a user vector of the user according to the characteristic data of each historical song, the characteristic data of the historical song list and the characteristic data of the user; determining target user preference of the user to songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph; and determining the recommended song list of the user according to the target user preference. The invention can improve the diversity of songs recommended in the recommended song list.

Description

Playlist recommendation method, device and readable storage medium

Technical field

The present invention relates to the technical field of content recommendation, and in particular to a song list recommendation method, device and readable storage medium.

Background technique

The existing playlist recommendation method is that the user inputs a search command, and the playlist recommendation device searches the song library for songs that match the search command according to the search command input by the user. If a song matching the search instruction is found, the found song will be used as a candidate song, and a recommended song list will be generated and recommended to the user based on the candidate song; if no song matching the search instruction is found, the song will be All songs included in the library are subjected to collaborative filtering processing to obtain multiple songs that meet the user's preferences as candidate songs, and then a recommended playlist is generated and recommended to the user based on the candidate songs. However, this playlist recommendation method is completely based on the user's search instructions. The songs recommended in the recommended playlist are relatively single and have poor diversity.

Contents of the invention

The main purpose of the present invention is to provide a playlist recommendation method, device and readable storage medium, aiming to improve the diversity of recommended songs in the recommended playlist.

In order to achieve the above object, the present invention provides a song list recommendation method. The song list recommendation method includes:

Obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user;

Determine the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user;

Determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

Determine the user's recommended playlist according to the target user's preferences.

Optionally, the step of determining the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user includes:

The first vector of each historical song is determined according to the characteristic data of each historical song, the second vector of the historical playlist is determined according to the characteristic data of the historical playlist, and the second vector of the historical playlist is determined according to the characteristic data of the user. User’s third vector;

Obtain the first weight value of each historical song to the historical play list, and determine the song list vector of the historical play list based on the first vector, the second vector and the first weight value;

Obtain the second weight value of the historical playlist for the user, and determine the user vector of the user based on the playlist vector, the third vector and the second weight value.

Optionally, the step of determining the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph includes:

Determine the user's candidate user preferences for songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

Obtain the preference score of the song under the user preference to be selected;

A target user preference for the song is determined based on the preference score.

Optionally, the step of obtaining the preference score of the song under the user preferences to be selected includes:

Obtain the song vector of the song and the preference vector of the candidate user preference;

Add the user vector and the song vector;

Similarity calculation is performed between the vector obtained after the addition and the preference vector of the candidate user preference, to obtain the preference score of the song under the candidate user preference.

Optionally, the step of obtaining the preference vector of the candidate user's preferences includes:

Determine the relationship corresponding to the user preferences to be selected according to the preset song knowledge graph;

The preference vector of the candidate user preference is determined according to the relationship vector of the relationship.

Optionally, the step of determining the user's recommended playlist according to the target user's preferences includes:

Obtain the user's recommendation score for the song under the target user's preferences;

The user's recommended playlist is determined based on the recommendation score.

Optionally, the step of obtaining the user's recommendation score for the song under the target user's preferences includes:

Determine the preference enhancement vector preferred by the target user according to the preference vector preferred by the target user;

determining a song enhancement vector for the song based on the song vector of the song;

The user's recommendation score for the song under the target user's preference is determined according to the user vector, the preference enhancement vector and the song enhancement vector.

In addition, in order to achieve the above object, the present invention also provides a song list recommendation device. The song list recommendation device includes:

The acquisition module is used to obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user;

A first determination module, configured to determine the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user;

a second determination module, configured to determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

A recommendation module, configured to determine the user's recommended playlist according to the target user's preferences.

In addition, to achieve the above object, the present invention also provides a playlist recommendation device. The playlist recommendation device includes a memory, a processor, and a playlist recommendation program stored in the memory and operable on the processor. , when the playlist recommendation program is executed by the processor, the steps of any of the above playlist recommendation methods are implemented.

In addition, in order to achieve the above object, the present invention also provides a computer-readable storage medium. A playlist recommendation program is stored on the computer-readable storage medium. When the playlist recommendation program is executed by a processor, any of the above items can be implemented. The steps of the playlist recommendation method.

The present invention proposes a song list recommendation method, device and readable storage medium. By obtaining the user's historical play list, the characteristic data of the historical song list, the characteristic data of each historical song included in the historical song list and the The user's characteristic data is determined based on the characteristic data of each historical song, the characteristic data of historical playlists and the user's characteristic data, and the user's response to the preset song knowledge graph is determined based on the user vector and the preset song knowledge graph. Target user preferences for songs, and determine the user's recommended playlist based on the target user preferences. This solution recommends playlists based on the song knowledge graph, which can accurately capture the user's preferences for songs. Recommending playlists based on the user's preferences for songs effectively increases the diversity of recommended songs in the recommended playlists.

Description of drawings

Figure 1 is a schematic diagram of the hardware architecture of a playlist recommendation device involved in an embodiment of the present invention;

Figure 2 is a schematic flow chart of the first embodiment of the playlist recommendation method of the present invention;

Figure 3 is a schematic flow chart of the second embodiment of the playlist recommendation method of the present invention;

Figure 4 is a schematic flow chart of the third embodiment of the playlist recommendation method of the present invention;

Figure 5 is a schematic module structure diagram of a playlist recommendation device according to an embodiment of the present invention.

The realization of the purpose, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

As an implementation solution, please refer to Figure 1. Figure 1 is a schematic diagram of the hardware architecture of a playlist recommendation device involved in an embodiment of the present invention. As shown in Figure 1, the playlist recommendation device may include a processor 101, such as a CPU. Memory 102, communication bus 103, wherein the communication bus 103 is used to realize connection communication between these modules.

The memory 102 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. As shown in Figure 1, the memory 102, which is a computer-readable storage medium, may include a playlist recommendation program; and the processor 101 may be used to call the playlist recommendation program stored in the memory 102 and perform the following operations:

Obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user;

Determine the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user;

Determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

Determine the user's recommended playlist according to the target user's preferences.

Further, the processor 101 can be used to call the playlist recommendation program stored in the memory 102, and perform the following operations:

The first vector of each historical song is determined according to the characteristic data of each historical song, the second vector of the historical playlist is determined according to the characteristic data of the historical playlist, and the second vector of the historical playlist is determined according to the characteristic data of the user. User’s third vector;

Obtain the first weight value of each historical song to the historical play list, and determine the song list vector of the historical play list based on the first vector, the second vector and the first weight value;

Obtain the second weight value of the historical playlist for the user, and determine the user vector of the user based on the playlist vector, the third vector and the second weight value.

Further, the processor 101 can be used to call the playlist recommendation program stored in the memory 102, and perform the following operations:

Determine the user's candidate user preferences for songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

Obtain the preference score of the song under the user preference to be selected;

A target user preference for the song is determined based on the preference score.

Further, the processor 101 can be used to call the playlist recommendation program stored in the memory 102, and perform the following operations:

Obtain the song vector of the song and the preference vector of the candidate user preference;

Add the user vector and the song vector;

Similarity calculation is performed between the vector obtained after the addition and the preference vector of the candidate user preference, to obtain the preference score of the song under the candidate user preference.

Further, the processor 101 can be used to call the playlist recommendation program stored in the memory 102, and perform the following operations:

Determine the relationship corresponding to the user preferences to be selected according to the preset song knowledge graph;

The preference vector of the candidate user preference is determined according to the relationship vector of the relationship.

Further, the processor 101 can be used to call the playlist recommendation program stored in the memory 102, and perform the following operations:

Obtain the user's recommendation score for the song under the target user's preferences;

The user's recommended playlist is determined based on the recommendation score.

Further, the processor 101 may be used to call the song list recommendation program stored in the memory 102, and perform the following operations:

Determine the preference enhancement vector preferred by the target user according to the preference vector preferred by the target user;

determining a song enhancement vector for the song based on the song vector of the song;

The user's recommendation score for the song under the target user's preference is determined according to the user vector, the preference enhancement vector and the song enhancement vector.

Currently, playlist recommendation is often based on the user inputting a search command. The playlist recommendation device searches the song library for songs that match the search command according to the search command input by the user. If a song matching the search instruction is found, the found song will be used as a candidate song, and a recommended song list will be generated and recommended to the user based on the candidate song; if no song matching the search instruction is found, the song will be All songs included in the library are subjected to collaborative filtering processing to obtain multiple songs that meet the user's preferences as candidate songs, and then a recommended playlist is generated and recommended to the user based on the candidate songs. Since this playlist recommendation method is completely based on the search instructions entered by the user, it cannot capture the user's inherent preference for selecting songs. However, the user is likely to like a song because of other fine-grained factors, such as the singer of the song. , genres, albums, etc. Therefore, the recommended songs in the recommended playlist obtained through the search instructions entered by the user are relatively single, even repeated, and the diversity is poor.

Based on the problems existing in the above-mentioned existing technologies, the song list recommendation method proposed by the present invention embeds user vectors and song knowledge graphs, and deduces the user preferences of songs based on the similarity scores on the song knowledge graph based on the user vectors, deducing the user's choice of songs. The user preferences behind it are used to recommend playlists to users based on the derived user preferences. User vectors and song knowledge graphs can capture users' intrinsic preferences for selecting songs, and recommending playlists based on user preferences can effectively increase the diversity of songs in the recommended playlists. Below, the song list recommendation method of the present invention will be further explained through specific embodiments.

Please refer to Figure 2. Figure 2 is a schematic flow chart of the first embodiment of the playlist recommendation method of the present invention. The playlist recommendation method includes:

Step S10, obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user;

In this embodiment, the execution subject is a playlist recommendation device, where the playlist recommendation device may be a terminal device, such as a computer, a mobile phone, a tablet, etc. Of course, in other embodiments, the playlist recommendation device may also be based on actual conditions. It needs to be determined, and this embodiment does not limit this.

In this embodiment, the playlist recommendation device obtains the user's historical playlist, and obtains the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user, where the historical songs It can be all the songs that the user has listened to, or it can be the songs that the user has listened to within a preset period, for example, the songs that the user has listened to in the past three months. Of course, the preset period can be set according to actual needs. In this embodiment There is no limit to this; the historical playlist is a collection of historical songs; the characteristic data of historical songs can include the name of the historical song, lyric information, lyricist, composer, singer, genre, album and other information; the characteristic data of the historical song list It may include information such as the number of songs contained in the playlist, song types, etc.; the user's characteristic data may include the user's identity information and other information associated with the user's identity information.

Step S20, determine the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user;

In this embodiment, after obtaining the characteristic data of historical songs, the characteristic data of historical playlists, and the characteristic data of users, the playlist recommendation device generates a User vector, where user vector refers to an embedded representation that reflects the user’s historical listening characteristics.

Specifically, the first vector of each historical song is determined according to the characteristic data of each historical song, the second vector of the historical playlist is determined according to the characteristic data of the historical song list, and the third vector of the user is determined according to the characteristic data of the user, wherein the th The first vector is the initialized embedded representation of the song, the second vector is the initialized embedded representation of the playlist, and the third vector is the initialized embedded representation of the user. By combining the characteristic data of historical songs, the characteristic data of historical playlists and the user's The feature data is input to the nonlinear network to obtain the first vector m, the second vector l ₀ and the third vector u ₀ of unified dimensions, where m, l ₀ , u ₀ , d is the embedding dimension.

After acquiring the first vector, the second vector and the third vector, the song list recommendation device acquires the first weight value of each historical song to the historical song list, and determines the history based on the first vector, the second vector and the first weight value. The playlist vector of the playlist, where the first weight value refers to the importance of historical songs to the historical playlist where they are located. For the historical playlist l, use Represents a collection of historical songs included in the historical song list l. The importance of different historical songs in a historical song list is different. For example, some historical songs may be representative songs of this historical song list. Use attention mechanism to assign weight values to different historical songs/> , weight value/> That is the first weight value, which represents the importance of historical song m in historical song list l. The first weight value can be determined using a forward neural network and normalized by the softmax function. The calculation process of the first weight value is as follows:

where m is the embedded representation of historical song m, is the weight matrix,/> is the bias vector, is the weight vector,/> is the offset value. Then aggregate the first vector of historical songs in the historical playlist l and combine it with the second vector of the historical playlist to obtain the songlist vector l of the historical playlist. The calculation process of the songlist vector of the historical playlist is as follows:

After obtaining the playlist vector of the historical playlist, the playlist recommendation device obtains the second weight value of the historical playlist for the user, and determines the user vector of the user based on the playlist vector, the third vector and the second weight value, where the The second weight value is the importance of the historical playlist to the user. use Represents a collection of historical playlists that user u has listened to. Different historical playlists have different importance to the user. Use attention mechanism to assign weight values to different historical playlists/> , weight value/> That is the second weight value, which represents the importance of the historical song list l to the user u. The second weight value can be determined using the forward neural network and normalized through the softmax function. The calculation process of the second weight value is as follows:

Among them, l is the song list vector of the historical song list l, is the weight matrix,/> is the bias vector, is the weight vector,/> is the offset value. Then the playlist vectors of the historical playlists listened to by user u are aggregated, and combined with the user's third vector, the user vector u of user u is obtained. The calculation process of user vector u is as follows:

Step S30: Determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

Step S40: Determine the user's recommended playlist according to the target user's preferences.

After the song list recommendation device obtains the user vector, it determines the user's target user preference for the songs in the preset song knowledge map based on the user vector and the preset song knowledge map, where the target user preference refers to the specific preference reason for the user to select the song.

The song list recommendation device is pre-constructed and embedded with a song knowledge graph. The song knowledge graph is a knowledge graph organized using the characteristic data of the song. The characteristic data of the song may include the name of the song, the lyricist of the song, the singer of the song, The album the song is in. The song list recommendation device can establish fact triples based on the song's characteristic data, and build and embed the song knowledge graph based on the fact triples. Represent the song knowledge graph as , where,/> Represents an entity set,/> Represents a relationship set. DISTMULT can be used to embed the knowledge graph. The score function of the fact triplet (h, r, t) is as follows:

in, Is the embedded representation of the head entity h,/> is the embedded representation of relation r,/> Indicates that the diagonal element is/> diagonal matrix of corresponding elements in ,/> is the embedded representation of the tail entity t. The loss function of the embedding part of the song knowledge graph adopts a threshold-based ranking loss. The calculation process of the loss function is as follows:

in, ,/> Including incorrect triples formed by randomly replacing the head entity or tail entity in the correct fact triplet,/> Controls the threshold between correct triples and incorrect triples.

It should be noted that while embedding the song knowledge graph, user preferences are modeled ,in , one-to-one correspondence between the relationships between entities in the song knowledge graph and the user's user preferences for songs. By establishing a one-to-one correspondence between the relationships in the song knowledge graph and user preferences, it is possible to deduce the user's preferences for songs based on the relationships in the song knowledge graph. Various possible user preferences, in this way, the preference vector of the user's user preference for the song can be determined through the relationship vector in the song knowledge graph.

Furthermore, based on the user vector and the song knowledge graph, the user's target user preference for the song can be derived from various possible user preferences for the song. After determining the target user preference for the song in the song knowledge graph, based on the user's target user preference for the song Prefers to recommend playlists to users.

In the technical solution provided by this embodiment, by obtaining the user's historical playlist, and respectively obtaining the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user, according to the characteristics of the historical song The characteristic data, the characteristic data of historical playlists and the user's characteristic data determine the user's user vector. Based on the user vector and the preset song knowledge graph, the user's target user preference for the songs in the preset song knowledge graph is determined. The target user preference is determined based on the target user's preference. Recommended playlists from users. This solution recommends playlists based on the song knowledge graph, which can accurately capture the user's inherent preference for songs. Recommending playlists based on the user's preferences for songs effectively increases the diversity of recommended songs in the recommended playlist.

Please refer to Figure 3. Figure 3 is a schematic flow chart of the second embodiment of the playlist recommendation method of the present invention. Based on the first embodiment, the above-mentioned steps of S30 include:

Step S31: Determine the user's candidate user preferences for the songs in the preset song knowledge map according to the user vector and the preset song knowledge map;

In this embodiment, after the song list recommendation device obtains the user vector, based on the one-to-one correspondence between the relationship in the song knowledge graph and the user preference, it determines the user's candidate songs in the song knowledge graph based on the user vector and the song knowledge graph. User preferences, where the user preferences to be selected are the possible reasons for the user to select the song. For example, the user preferences to be selected may include the lyricist, singer, genre, and album of the song.

Step S32: Obtain the preference score of the song under the preference of the user to be selected;

In this embodiment, after determining the user preference of the song to be selected, the song list recommendation device obtains the preference score of the song under the user preference of the song to be selected. Specifically, the song list recommendation device obtains the song vector of the song through the preset song knowledge graph and determines the relationship corresponding to the preference of the candidate user according to the preset song knowledge graph, and determines the preference of the candidate user according to the relationship vector of the relationship corresponding to the preference of the candidate user. The preference vector is then added to the user vector and the song vector, and the similarity between the added vector and the preference vector of the user to be selected is calculated to obtain the preference score of the song under the preference of the user to be selected. The preference score is calculated as follows:

where u is the embedded representation of user u, m is the embedded representation of song m, p is the embedded representation of preference p, Use dot product operation.

Optionally, the playlist recommendation device can use Gumbel SoftMax to discretely sample user preferences. This method utilizes the re-parameterization technique of backpropagation so that continuous gradients of model parameters can be calculated during end-to-end training. ST-Gumbel-SoftMax approximates a one-hot vector from a multi-class distribution. Assume that the probability of belonging to class p in the P-way classification distribution can be defined as:

After that, sample a one-hot vector from the above distribution as follows:

in, is Gumbel noise, u is generated by a specific noise distribution.

Step S33: Determine the target user preference of the song according to the preference score.

In this embodiment, after determining the preference scores of the songs under the preferences of the users to be selected, the song list recommendation device can sort the songs according to the partial scores of the songs under the preferences of the users to be selected, and select the songs with the highest preference scores according to the preferences of the users to be selected. Determine target user preferences for songs. By repeating the above process, the target user preferences of all songs in the song knowledge graph can be determined.

In the technical solution provided by this embodiment, the user's candidate user preference for the song in the preset song knowledge map is determined based on the user vector and the preset song knowledge map, and then the preference score of the song under the candidate user preference is obtained. According to the preference The score determines the target user preference of the song. This solution can accurately deduce the target user preference of the song selected by the user through the user vector and the preset song knowledge map, and then recommends the song list to the user based on the target user preference, which can improve the recommended song list. Variety of songs.

Please refer to Figure 4. Figure 4 is a schematic flow chart of the third embodiment of the playlist recommendation method of the present invention. Based on the first embodiment, the above-mentioned steps of S40 include:

Step S41: Obtain the user's recommendation score for the song under the target user's preferences;

In this embodiment, after determining the target user preferences of all songs in the preset song knowledge map, the song list recommendation device obtains the user's recommendation score for the song under its target user preference, and calculates the user's recommendation score for the song under its target user preference. The recommendation score below can know how much the user likes the song, and then determine the possibility of recommending the song to the user.

Specifically, the preference enhancement vector of the target user's preference is determined based on the preference vector of the target user's preference, the song enhancement vector of the song is determined based on the song's song vector, and the user's preference for the song in the target user's preference is determined based on the user vector, preference enhancement vector and song enhancement vector. recommendation score below. The calculation process of recommendation score is as follows:

Calculate preference p, enhancement vector of song m ,/> :

Among them, r is the embedded representation of the relationship r corresponding to the preference p, and e is the embedded representation of the entity e corresponding to the song m. You can refer to the method of calculating the fact triple score in DISTMULT to calculate the recommendation score of user u for music m under the target user preference p:

in, Indicates that the diagonal element is/> diagonal matrix of corresponding elements in . The loss function of the recommendation module is:

Step S42: Determine the user's recommended playlist based on the recommendation score.

In this embodiment, the recommendation scores of all songs in the preset song knowledge graph can be determined through the above method, that is, the user's recommendation scores for all songs under their corresponding target user preferences. The song list recommendation device obtains the song knowledge graph. After the recommendation scores of all songs are obtained, the recommendation scores can be sorted, and the preset number of songs with the highest recommendation scores will be used as the songs to be recommended, and the songs to be recommended will be composed into a recommended play list and recommended to the user. The preset number can be based on Actual needs are determined, and this embodiment does not limit this.

Optionally, the playlist recommendation device can also jointly optimize the loss function of the knowledge graph embedding module and the loss function of the playlist recommendation module. The optimization method is as follows:

in, It is a hyperparameter that balances the song knowledge graph embedding module and the recommendation module.

In the technical solution provided by this embodiment, the user's recommendation score for songs under the target user's preferences is obtained, and the user's recommended song list is determined based on the recommendation score. This solution determines the recommendation score through the user's target user preference for songs, and recommends playlists based on the recommendation scores, which improves the diversity of recommended songs in the recommended playlist.

Based on the above embodiments, please refer to Figure 5. The present invention also provides a song list recommendation device. The song list recommendation device includes:

The acquisition module 100 is used to obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user;

The first determination module 200 is configured to determine the user vector of the user based on the characteristic data of each historical song, the characteristic data of the historical playlist, and the characteristic data of the user;

The second determination module 300 is configured to determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph;

The recommendation module 400 is configured to determine the user's recommended playlist according to the target user's preferences.

Based on the above embodiments, the present invention also provides a playlist recommendation device. The playlist recommendation device may include a memory, a processor, and a playlist recommendation program stored in the memory and runable on the processor. The above processing When the device executes the above playlist recommendation program, the steps of the playlist recommendation method described in any of the above embodiments are implemented.

Based on the above embodiments, the present invention also provides a computer-readable storage medium on which a playlist recommendation program is stored. When the above playlist recommendation program is executed by a processor, the playlist recommendation method as described in any of the above embodiments is implemented. A step of.

It should be noted that, as used herein, the terms "include", "comprising" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or system that includes a list of elements not only includes those elements, but It also includes other elements not expressly listed or that are inherent to the process, method, article or system. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

The above serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM) as mentioned above. , magnetic disk, optical disk), including several instructions to cause a terminal device (which can be a smart TV, a mobile phone, a computer, etc.) to execute the method described in various embodiments of the present invention.

The above are only preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the description and drawings of the present invention may be directly or indirectly used in other related technical fields. , are all similarly included in the scope of patent protection of the present invention.

Claims (9)

1. A playlist recommendation method, characterized in that the playlist recommendation method includes: Obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user; The first vector of each historical song is determined according to the characteristic data of each historical song, the second vector of the historical playlist is determined according to the characteristic data of the historical playlist, and the second vector of the historical playlist is determined according to the characteristic data of the user. User’s third vector; Obtain the first weight value of each historical song to the historical play list, and determine the song list vector of the historical play list based on the first vector, the second vector and the first weight value; Obtain the second weight value of the historical playlist for the user, and determine the user vector of the user based on the playlist vector, the third vector and the second weight value; Determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph; Determine the user's recommended playlist according to the target user's preferences. 2. The song list recommendation method according to claim 1, wherein the target user preference for the songs in the preset song knowledge graph is determined by the user according to the user vector and the preset song knowledge graph. The steps include: Determine the user's candidate user preferences for songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph; Obtain the preference score of the song under the user preference to be selected; A target user preference for the song is determined based on the preference score. 3. The song list recommendation method according to claim 2, wherein the step of obtaining the preference score of the song under the preference of the candidate user includes: Obtain the song vector of the song and the preference vector of the candidate user preference; Add the user vector and the song vector; Similarity calculation is performed between the vector obtained after the addition and the preference vector of the candidate user preference, to obtain the preference score of the song under the candidate user preference. 4. The playlist recommendation method according to claim 3, wherein the step of obtaining the preference vector of the candidate user's preferences includes: Determine the relationship corresponding to the user preferences to be selected according to the preset song knowledge graph; The preference vector of the candidate user preference is determined according to the relationship vector of the relationship. 5. The playlist recommendation method according to claim 1, wherein the step of determining the user's recommended playlist according to the target user's preferences includes: Obtain the user's recommendation score for the song under the target user's preferences; The user's recommended playlist is determined based on the recommendation score. 6. The song list recommendation method according to claim 5, wherein the step of obtaining the user's recommendation score for the song under the target user's preference includes: Determine the preference enhancement vector preferred by the target user according to the preference vector preferred by the target user; determining a song enhancement vector for the song based on the song vector of the song; The user's recommendation score for the song under the target user's preference is determined according to the user vector, the preference enhancement vector and the song enhancement vector. 7. A song list recommendation device, characterized in that the song list recommendation device includes: The acquisition module is used to obtain the user's historical playlist, and obtain the characteristic data of the historical playlist, the characteristic data of each historical song included in the historical playlist, and the characteristic data of the user; The first determination module is configured to determine the first vector of each historical song according to the characteristic data of each historical song, determine the second vector of the historical song list according to the characteristic data of the historical song list, and determine the second vector of the historical song list according to the characteristic data of the historical song list. Characteristic data of the user determines a third vector of the user; Obtain the first weight value of each historical song to the historical play list, and determine the song list vector of the historical play list based on the first vector, the second vector and the first weight value; Obtain the second weight value of the historical playlist for the user, and determine the user vector of the user based on the playlist vector, the third vector and the second weight value; a second determination module, configured to determine the user's target user preference for the songs in the preset song knowledge graph according to the user vector and the preset song knowledge graph; A recommendation module, configured to determine the user's recommended playlist according to the target user's preferences. 8. A playlist recommendation device, characterized in that the playlist recommendation device includes a memory, a processor, and a playlist recommendation program stored in the memory and executable on the processor, and the playlist recommendation program When the recommendation program is executed by the processor, the steps of the playlist recommendation method according to any one of claims 1-6 are implemented. 9. A computer-readable storage medium, characterized in that a playlist recommendation program is stored on the computer-readable storage medium, and when the playlist recommendation program is executed by a processor, it implements any one of claims 1-6 The steps of the playlist recommendation method described in the item.