WO2020151189A1 - Song recommendation method and system, terminal and storage medium - Google Patents

Abstract

A song recommendation method and system, a terminal and a storage medium. The method comprises: acquiring attribute parameters of a first song (S101), wherein the attribute parameters comprise at least one of a song collection parameter, a song search parameter and a song playing parameter; calculating a preference degree of the first song according to the attribute parameters (S102); calculating, according to the preference degree of the first song, values of favorability of a user for different song characteristics, and calculating a preference degree of a second song according to the values of favorability of the user for different song characteristics and weights of the different song characteristics for the second song (S103); and outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song (S104). The method can greatly improve the processing efficiency on the basis of providing the fitting accuracy of song recommendation, is particularly suitable in a vehicle scenario, and can be widely applied to the field of recommendations.

Description

Song recommendation method, system, terminal and storage medium Technical field

The present invention relates to data processing technology, in particular to a song recommendation method, system, terminal and storage medium.

Background technique

In order to understand lack, relieve the dull atmosphere in the car or satisfy personal hobbies, car owners often play music while driving. Usually, car owners will pre-download and store the songs they want to listen to on the in-vehicle system or mobile device. When music needs to be played, the downloaded song files will be looped and/or randomly played to ensure that the played songs can be played. It meets the personal preferences of the car owner, but the number of songs to be played is limited, and it needs to be downloaded and stored in advance, and the operation convenience is low. Therefore, in order to solve this problem, a song recommendation algorithm is designed to recommend the corresponding song to the owner according to the song that the owner listens to. This can meet the owner's personal preference, without pre-downloading and increasing the number of songs, thereby improving the operation Convenience and number of songs that can be played. However, currently commonly used mainstream song recommendation algorithms, such as collaborative filtering, purely regular stack recommendation scoring mechanism sorting, etc., have a low degree of fit between the recommendation results and the songs that users love, and through the establishment of songs or playlists To realize song recommendation based on the similarity relationship matrix, there will be a problem of complicated calculation and lower processing efficiency.

Summary of the invention

In view of this, the purpose of the embodiments of the present invention is to provide a song recommendation method, system, terminal, and storage medium to implement fast and accurate song recommendation.

In the first aspect, an embodiment of the present invention provides a song recommendation method, including the following steps:

Acquiring attribute parameters of the first song; the attribute parameters include at least one of song collection parameters, song search parameters, and song playback parameters;

Calculate the preference of the first song according to the attribute parameter;

According to the preference of the first song, the user’s preference for different song characteristics is calculated, and the second song is calculated according to the user’s preference for different song characteristics and the weight of different song characteristics on the second song Preference

According to the preference degree of the first song and/or the preference degree of the second song, the song recommendation result is output.

In the second aspect, an embodiment of the present invention provides a song recommendation system, including:

The first obtaining unit is configured to obtain attribute parameters of the first song; the attribute parameters include at least one of song collection parameters, song search parameters, and song playback parameters;

The first calculation unit is configured to calculate the preference degree of the first song according to the attribute parameter;

The second calculation unit is used to calculate the user’s preference for different song characteristics according to the preference of the first song, and according to the user’s preference for different song characteristics and the preference of different song characteristics for the second song Weight, calculate the preference of the second song;

The recommendation unit is configured to output song recommendation results according to the preference degree of the first song and/or the preference degree of the second song.

In a third aspect, an embodiment of the present invention provides a terminal, and the device includes:

At least one processor;

At least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the song recommendation method.

In a fourth aspect, an embodiment of the present invention provides a storage medium in which instructions executable by a processor are stored, and the instructions executable by the processor are used to execute the song recommendation method when executed by the processor.

One or more technical solutions in the above embodiments of the present invention have the following advantages: the embodiment of the present invention calculates the preference of the first song by using the attribute parameters of the first song, and the attribute parameters include song collection behavior parameters, song At least one of search behavior parameters and song playing behavior parameters, and then calculate the user’s preference for different song characteristics according to the preference of the first song, and according to the user’s preference for different song characteristics and different The weight of the song characteristic to the second song is calculated, the preference of the second song is calculated, and then the song recommendation result is output according to the preference of the first song and/or the preference of the second song. It can be seen that the song recommendation method of the embodiment of the present invention combines the user's operating behavior on the song, the user's preference for the song characteristics, and the weight of the song characteristics on the song to achieve song recommendation, wherein the user's operating behavior on the song includes There are song collection, search and/or playing behaviors. Therefore, compared with traditional recommendation techniques, the embodiment of the present invention can greatly improve the fit between the final output recommendation result and the user's favorite song, and at the same time, the preference for the second song It does not need to construct a similarity relationship matrix between songs or playlists, which can greatly improve the processing efficiency on the basis of the accuracy of the song recommendation.

Description of the drawings

FIG. 1 is a schematic diagram of the steps of a specific embodiment of a song recommendation method according to an embodiment of the present invention;

FIG. 2 is a schematic flowchart of steps in another specific embodiment of a song recommendation method according to an embodiment of the present invention;

Figure 3 is a structural block diagram of a song recommendation system according to an embodiment of the present invention;

Fig. 4 is a structural block diagram of a terminal according to an embodiment of the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the drawings and specific embodiments. For the step numbers in the following embodiments, they are set only for the convenience of elaboration, and there is no limitation on the order between the steps. The execution order of the steps in the embodiments can be adapted according to the understanding of those skilled in the art. Sexual adjustment.

As shown in FIG. 1, an embodiment of the present invention provides a song recommendation method, which includes the following steps.

S101. Obtain attribute parameters of the first song; the attribute parameters include at least one user behavior parameter among song collection parameters, song search parameters, and song playback parameters.

Specifically, for the first song, it refers to a song that has any behavior of collecting, searching, and playing, that is, the song that has been collected, searched, and/or played by the user is an embodiment of the present invention Therefore, the attribute parameter of the first song may include at least one user behavior parameter among song collection parameters, song search parameters, and song playback parameters.

For song collection parameters, it may include but is not limited to the number of song collections, song collection time, song collection frequency, etc. Generally, the more song collections and/or the higher the collection frequency, the easier it is for users to like the song mood. Wherein, the number of song collection times is recorded by clicking the number of times the song collection button is clicked. For example, if the song collection button is clicked once, the number of song collection times is increased by 1; the song collection time is measured by the song collection button Click the time to record. For example, the time when the song collection button is clicked is the song collection time; the song collection frequency is obtained by counting the number of song collections within a preset time period. The higher the number of collections of songs in a time period, the higher the frequency of collection of songs. Among them, for the preset time period, it is a time period that can be manually selected and set. You can also press from a number of song collection times Two song collection times need to be selected, and the time period between the two song collection times is the preset time period.

For the song search parameter, it may include but is not limited to the number of song searches, song search frequency, song search time, etc. Generally, the more song searches and/or the higher the search frequency, the more the user likes the song. Wherein, the number of search times of the song is counted by the number of search records of the song. For example, if the searched record of song A is 1, then the number of search times of the song is increased by 1, and the search record of the song is usually Triggered by the user entering the song name into the search box and clicking the search button; the song search time is recorded by the click time of the song search button, such as the time when the song search button is clicked is the song search time; The song search frequency is obtained by counting the number of song searches within a preset time period. The higher the number of song searches in the preset time period, the higher the song search frequency. The preset time period can be a time period set by manual selection, or two song search times selected from a number of song search times as needed, and the time period between the two song search times That is the preset time period.

For the song playback parameters, it may include, but is not limited to, total song playback time, song playback frequency, song playback times, etc. Generally, the longer the total song playback time, the higher the song playback frequency and/or the more song playback times. , It means that the user likes the song more. Wherein, the total playing time of the song can be calculated from the number of times the song is played and the duration of the song. For example, if the song has a duration of 4 minutes and a total of 10 times have been played completely, the playing time of the song will be 40 minutes. If the song is not played completely, it will be superimposed according to the duration of the i-th play. For example, the duration of the song is 4 minutes, and it has been completely played 9 times. At the 10th time, the song only played 2 minutes. The playing time of the song is 38 minutes. The statistical methods of the number of times of song playback may include: when song A is played for more than a preset time period, it means that the number of times of song A is played plus 1; when song A is played repeatedly once, then the number of times of song A is played plus 1; During the playback of song A, the playback progress bar of song A is pulled back to the starting point or other preset playback points to restart the playback of the song audio segment, and the playback exceeds the preset duration or the playback reaches the end of the single. At this time, song A The number of plays is also increased by 1. The song playing frequency is obtained by counting the number of times the song is played in a preset time period. The higher the song playing times in the preset time period, the higher the song playing frequency. The preset time period (usually in units of days), which can be a time period set manually, or from the number of song playbacks, the i-th song and the j-th song can be selected as needed. The starting time point corresponding to the i-th song playing and the starting time point corresponding to the j-th song playing, the time period between the two is the preset time period. Among them, in the process of collecting and acquiring song playback parameters, the song playback buttons such as the song start button, the song end playback button, the song progress bar pull button, and the song loop playback button are usually monitored. The required song playback parameters.

For the above song collection button, song search button and/or song play button, it can be a virtual button displayed in the client program, or a physical button (ie, physical button) set in the terminal device, which can be based on actual conditions. Need to design, here is not too limited.

S102. Calculate the preference of the first song according to the attribute parameter.

Specifically, since the attribute parameters include song collection parameters, song search parameters, and/or song playback parameters, all of which reflect the user’s preference for the song, the song collection parameters, song search parameters, and/or song playback The preference degree of the first song obtained after the parameter calculation process can accurately fit the user's preference for the first song. Among them, for the preference of the first song, the calculation methods that can be used may include, but are not limited to: directly performing calculations on the song collection parameters, song search parameters, and/or song playback parameters to obtain the first song. Song preference degree; or, after the song collection parameters, song search parameters and/or song playback parameters are converted into corresponding values according to the preset unit, the calculation process is performed to obtain the preference degree of the first song; or, After the song collection parameters, song search parameters, and/or song playback parameters are weighted, arithmetic processing is performed. For the calculations described, it may include but is not limited to addition, subtraction, multiplication, division, exponent, logarithmic, etc. Therefore, when calculating the preference of the first song, it can be based on actual needs. To select the calculation method, there is no excessive limitation in this embodiment.

S103. Calculate the user's preference for different song characteristics according to the preference of the first song, and calculate the second song based on the user's preference for different song characteristics and the weight of different song characteristics on the second song. 2. The preference of the song. For the second song, it refers to all songs except the first song, that is, the second song is a song without song collection behavior, song search behavior, and song playback behavior. As for the song characteristics, it may include sentimental, Chinese, English, girl group and other characteristics.

Specifically, since the preference of the first song is calculated according to the above-mentioned user behavior parameter, and the user behavior parameter reflects the user’s preference for the song, the preference of several first songs is used , Can calculate the user's preference for different song characteristics. Among them, the method of calculating the user’s preference for different song characteristics may include: taking the preference of the first song as the weight coefficient of the song characteristics of the first song. For example, the song characteristics of the first song A include characteristics 1, characteristics 2. And feature 3, and then multiply the preference of song A as the weighting coefficient by the three feature values of feature 1, feature 2, and feature 3. Then do the same for each song feature of the first song, and then Perform numerical statistics on the characteristics of each type of song. In this way, according to the numerical value corresponding to the song characteristic, the user's preference for different song characteristics can be obtained. Generally, the greater the numerical value corresponding to the song characteristic, the user's preference for the song characteristic The higher the degree of preference, the greater the value of the degree of preference.

Then, combining the user’s preference for different song characteristics and the weight of several song characteristics contained in the second song in the second song (that is, the weight of different song characteristics on the second song), the user can be calculated The degree of preference for the second song, that is, the degree of preference for the second song. Wherein, the weights of several song characteristics of the second song may be pre-stored in a server or other devices as attributes of the second song, or calculated according to the audio characteristics or song tags of the second song.

S104: Output a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song.

Specifically, according to the preference degree of the first song and/or the preference degree of the second song, that is, the preference degree of all songs, the song recommendation result is output according to the recommendation rule, where the recommendation rule may be Contains: sort all songs from largest to smallest according to the value of preference, and then select the top n songs as the song recommendation result. At this time, the plurality of songs include the first song and/or The second song; or, according to the preference value, sort the second songs from largest to smallest, and then select all the second songs from the top n songs as the song recommendation results. Further, if you select The number of all second songs out does not meet the preset number, such as 20, then the first song is selected from the top n songs, so that the number of songs included in the song recommendation result is The preset number; or, for the first song whose play, search, and/or collection behavior has been generated more than a preset time period (such as 1 month), such as the play, search, and /Or the collection behavior was generated 1 month ago, then only the first songs can be sorted according to their preference, and then several first songs ranked in the top n are selected as the song recommendation result. It can be seen that the recommended rules can be selected according to actual needs, and are not too limited in this embodiment.

It can be seen from the above that the embodiment of the present invention uses the song collection behavior parameter, the song search behavior parameter and/or the song playback behavior parameter of the first song to calculate the preference of the first song, and then the preference of the first song is used to calculate the preference of the first song. Calculate the user’s preference for the song characteristic, and calculate the preference of the second song according to the user’s preference for the song characteristic and the weight of the song characteristic on the second song, so as to calculate the preference of the second song according to the first song and/or the second song It can be seen that the embodiment of the present invention combines the user's operation behavior on the song (including song collection behavior, search behavior, and playback behavior) to achieve the preference degree calculation of the first song, and uses the first song Calculate the user’s preference for song characteristics and combine the weights of different song characteristics on the second song to calculate the second song’s preference. Therefore, compared to traditional collaborative filtering and pure rule stacking Recommendation, the results recommended by the solution of the embodiment of the present invention are more closely related to the user’s favorite songs, that is to say, the accuracy of the recommended results is higher, and there is no need to construct songs or songs during the calculation of song recommendations. The similarity relationship matrix between the singles can greatly improve the processing efficiency on the basis of the matching accuracy of the song recommendation.

Further as a preferred embodiment of the method, the step of outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song preferably includes:

According to the preference value of the second song, sort a number of second songs from largest to smallest, and then select a number of second songs in the top n as the song recommendation result. It can be seen that by selecting such a recommendation rule, not only can the user's favorite demand for songs be well satisfied, but also the freshness of the user's listening to songs can be improved. Further, if the number of the selected second songs is less than the preset number threshold, sort a number of first songs from largest to smallest, and then select a number of first songs in the top n1 to make the selection The number of songs out is equal to the preset number threshold.

Further as a preferred embodiment of the method, the user’s preference for different song characteristics is calculated according to the preference of the first song, and the preference value of the user for different song characteristics and the pairs of different song characteristics The weight of the second song, the step S103 of calculating the preference of the second song, includes:

S1031. Put the preferences of different users for different first songs into a song preference matrix, where the number of rows in the song preference matrix represents the number of users, and the number of columns in the song preference matrix represents the number of songs.

Specifically, first, the number of users is used as the number of rows and the number of songs is used as the number of columns to construct a song preference matrix, and then the calculated first preferences of different users for different first songs are put into the song preference. The corresponding element position of the matrix. For the preference matrix, it can be as shown in Table 1 below:

Table 1

Song preference	Song 1	Song 2	Song 3	Song 4	Song 5	Song 6	Song 7	Song 8
User 1	To	To	3	To	To	To	5	To
User 2	To	To	To	2	2	To	To	To
User 3	4	To	To	To	To	To	To	4
User 4	3	1	To	2	To	3	To	To
User 5	To	To	To	To	5	To	To	4
User 6	To	To	5	6	To	To	1	To

It can be seen that the above table 1 contains the preferences of 6 different users for 8 different songs, that is, the song preference matrix R′ _ui , u=1, 2, 3,..., m1, i=1, 2, 3...., n1, m1 is the total number of rows in the song preference matrix, n1 is the total number of columns in the song preference matrix, that is to say, m1 is the total number of users, n1 is the total number of songs, and in this implementation In the example, the total number of users is 6, and the total number of songs is 8. Therefore, the song preference matrix r′ _ui represents the preference of the u-th user for the i song, for example, r′ ₄₁ represents the preference of the fourth user for the first song. Among them, if song i is the first song for user u, then user u's user behavior parameters for song i (song collection parameters, search parameters, playback parameters) are used to calculate song i’s preference, and then The calculated preference is placed in the position (u, i). For example, song 3 is the first song for user 1, then after user 1’s user behavior parameters for song 3 are used to calculate the preference, it will be calculated The preference degree 3 out is placed in the position (1, 3), at this time r′ ₁₃ = 3; if the song i is the second song for the user u, then the element at the position (u, i) is Empty, or set the value of the element at this position to 0.

S1032. After matrix decomposition is performed on the song preference matrix, the preference matrix for different song characteristics of different users and the weight matrix of different song characteristics for different songs are obtained, wherein the weight matrix of different song characteristics for different songs is Contains the weights of different song characteristics to different second songs.

Specifically, for the different users’ preference matrix Q for different song characteristics, the number of rows represents the number of users, and the number of columns represents the number of song characteristics, so q _uf represents the characteristics of the uth user for the fth song Like the value of. The form of the matrix Q can be as shown in Table 2 below:

Table 2

To	Song Features 1	Song Features 2	Song Features 3
User 1	q 11	q 12	q 13
User 2	q 21	q 22	q 23

User 3	q 31	q 32	q 33
User 4	q 41	q 42	q 34
User 5	q 51	q 52	q 35
User 6	q 61	q 62	q 36

As for the weight matrix P ^{T of} different song characteristics for different songs, the number of columns indicates the number of songs, and the number of rows indicates the number of song characteristics, so p ^T _fi indicates that the f-th song characteristic is in the i-th song The weight occupied. The form of the matrix P ^T is shown in Table 3 below:

table 3

To	Song 1	Song 2	Song 3	Song 4	Song 5	Song 6	Song 7	Song 8
Song Features 1	p T 11	p T 12	p T 13	p T 14	p T 15	p T 16	p T 17	p T 18
Song Features 2	p T 21	p T 22	p T 23	p T 24	p T 25	p T 26	p T 27	p T 28
Song Features 3	p T 31	p T 32	p T 33	p T 34	p T 35	p T 36	p T 37	p T 38

It can be seen that in this embodiment, f=1, 2, 3; and the weight matrix P ^{T of} different song characteristics to the song is the transposed matrix of the matrix P, that is, the number of rows of the matrix P represents the song The number of columns represents the number of song characteristics, that is, the element value contained in the matrix P is P _if .

S1033. After multiplying the favorite degree matrix Q and the weight matrix P ^{T to} obtain a song score matrix R. At this time, the song rating matrix R includes all users' preferences for all songs, that is, the song rating matrix R includes the preference of the first song and the preference of the second song, so , The preference of the second song can be obtained from the song score matrix R. In fact, the song preference matrix R′ and the song rating matrix R have the same matrix form, except that the matrix R′ only contains the preferences of different users for different first songs calculated in step S102, and the matrix R In contains the preferences of different users for the first song and the second song obtained by multiplying Q and P ^T. Among them, the preference in matrix R is essentially the predicted value, and R′ contains The preference of the first song is the actual value.

S1034. Obtain the preference degree of the second song from the song score matrix.

It can be seen that the purpose of this embodiment is to calculate the song rating matrix R (R=QP ^T ) for different songs by different users. Therefore, the UV decomposition of the matrix is used to decompose the matrix R′ to obtain two low-dimensional Matrix Q and P ^T , and then use the product of the two matrices P ^T and Q to estimate the actual score matrix, that is, first solve the following objective function:

Let ∑(R′-QP ^T ) ² be the minimum value, and use the gradient descent method to get the estimated values of the elements contained in the matrix Q and P, and then use the estimated values of P and Q to calculate the matrix R All values. It can be seen that the automatic clustering of songs by user behavior is realized by using the principle of LFM (latent factor model, implicit semantic model), and the user’s preference for the second song is obtained through the element values contained in the matrix P and Q, and the calculation is fast And it is accurate, and this step is achieved through the LFM principle, without considering the angle of classification, the second song’s preference results are all based on the statistical automatic clustering of user behavior; at the same time, there is no need to consider the classification granularity, because the LFM is set The final classification number (that is, the number of users, the number of songs, the number of song characteristics) can control the granularity. The larger the classification number, the finer the granularity; and for each song, it is not clearly classified into a certain category (song characteristics ), but to calculate the probability that it belongs to each category, which belongs to soft classification. It can be seen that using this step to realize the preference degree of the second song can not only further perform the calculation process more quickly, but also reduce the error RMSE and improve the accuracy of preference degree calculation.

Further as a preferred embodiment of the present invention, the step S104 of outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song includes:

S10411. Acquire vehicle-mounted perception information. Wherein, the vehicle-mounted perception information includes, but is not limited to, temperature parameters in and/or in the car, humidity parameters in the car, vehicle speed, special weather, festivals, and the state of the user (such as the fatigue state of driving).

S10412. Obtain a song tag matching the vehicle-mounted perception information.

Specifically, when the temperature parameters inside and/or outside the car are higher than the preset temperature threshold, it means that the temperature is higher at this time, and then the song tags that match the car’s perception information at this time are obtained, such as "burning" and "heating". ", "fire", etc.; or, when the vehicle speed is continuously lower than the preset speed threshold for a preset period of time, it means that the vehicle continues to drive at a low speed, then the song tag that matches the vehicle perception information at this time is obtained, such as "light music ", "Quiet", etc. It can be seen that the essence of step S10412 is to determine the current vehicle environment or vehicle condition based on the acquired vehicle perception information, and then obtain the corresponding song tag according to the determined vehicle environment or vehicle condition.

S10413. When it is determined that the first song has the song tag, increase the preference of the first song, and/or when it is determined that the second song has the song tag, increase the The preference of the second song. That is, when the song tag of the first song and/or the second song contains "burn", "hot", "fire" or "light music", "quiet", etc., increase the first song and/or Or the preference of the second song.

S10414: Output a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song. It can be seen that by using vehicle-mounted perception information to increase the preference of songs, the output song recommendation results can not only fit the user's favorite direction, but also fit the current vehicle environment, which greatly improves the user's listening experience , And can also remind users of the vehicle environment and/or conditions by way of music playback.

Alternatively, as a further preferred embodiment of the method, the step S104 of outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song includes:

S10421: After summing the preferences of several first songs included in the playlist and/or the preferences of several second songs included in the playlist, the recommendation degree of the playlist is obtained;

S10422: Output song recommendation results according to the recommendation degree of the playlist.

It can be seen that in this embodiment, the song preference degree included in the song list is used to obtain the song list recommendation degree, and then the song list is output as the recommendation result, so that the recommendation is performed in the way of song batch recommendation, which can further speed up the processing. Efficiency, and the user can get multiple song recommendations through the selection of the playlist, which not only increases the user’s operational interaction experience, but also reduces the user’s operation and improves the operational convenience. This is especially suitable for users who are difficult to communicate with the client frequently. Car scene of interactive operation.

Further as a preferred embodiment of the method, the step S10422 of outputting the song recommendation result according to the recommendation degree of the playlist includes:

S104221. Acquire vehicle-mounted perception information. Wherein, the vehicle-mounted perception information includes, but is not limited to, temperature parameters in and/or in the car, humidity parameters in the car, vehicle speed, special weather, festivals, and the state of the user (such as the fatigue state of driving).

S104222. Acquire a playlist tag matching the vehicle-mounted perception information.

Specifically, when the temperature parameters inside and/or outside the car are higher than the preset temperature threshold, it means that the temperature is higher at this time, and then the playlist tags that match the vehicle perception information at this time are obtained, such as "burn", " "Hot", "Fire", etc.; or, when the vehicle speed is continuously lower than the preset speed threshold for a preset period of time, it means that the vehicle continues to drive at a low speed, then the playlist tag that matches the vehicle perception information at this time is obtained, such as "Light music", "Quiet", etc. It can be seen that the essence of step S104222 is to determine the current vehicle environment or vehicle condition based on the acquired vehicle perception information, and then obtain the corresponding playlist tag according to the determined vehicle environment or vehicle condition.

S104223: When it is determined that the playlist has the playlist tag, increase the recommendation degree of the playlist. That is, when the playlist tags of the playlist include "burn", "hot", "fire", "light music", "quiet", etc., the recommendation degree of the playlist is increased. As for the preference of the first song and/or the second song included in the playlist, they are adjusted based on the vehicle-mounted perception information, which can be implemented by using the above steps S10411 to S10413.

S104224: Output a song recommendation result according to the recommendation degree of the playlist.

It can be seen that by using vehicle-mounted perception information to increase the recommendation of the playlist, the output song recommendation results can not only fit the user's favorite direction, but also fit the current vehicle environment, which greatly improves the user's listening experience It can also remind users of the vehicle environment and/or conditions in the form of music playback.

Further as a preferred embodiment of the present invention, it also includes the following steps:

S10423. Obtain the popularity of the playlist, the time period since the update of the playlist, and/or the number of times of interest tag hits; for the popularity of the playlist, it can be obtained through statistics of the playlist and/or the amount of audition. The time period since the update of the playlist refers to the time period between the update time of the playlist and the current time. For the number of hits of the interest tag, it refers to the hits of the playlist tag of the playlist. The number of preset user interest tags selected by the user, for example, the playlist tags have A, B, C, D, and the preset user interest tags selected by the user are A, C, E, F, at this time interest The tags hit two, A and C, that is to say, the number of hits of the interest tag is 2

S10424. Adjust the recommendation degree of the playlist according to the popularity of the playlist, the time period since the update of the playlist and/or the number of hits of the interest tag; for the steps S10423 and S10424, they are set in step Before S10422;

Wherein, the popularity of the playlist is in a positive relationship with the recommendation degree of the playlist, the time period since the update of the playlist is in an inverse proportional relationship with the recommendation degree of the playlist, and the number of hits of the interest tag is proportional to the song recommendation. The recommendation degree of a single is proportional. It can be seen that in this embodiment, the influence factors such as the popularity of the playlist, the time since the update of the playlist, and/or the number of times of interest tag hits are added to adjust the recommendation degree of the playlist, which can further make the recommended songs more suitable for users. At the same time as the song likeness, it can also recommend newer, more popular songs that are more in line with the user's likeness to users, further improving the user's sense of listening experience.

Further as a preferred embodiment of the method, the song collection parameter includes the time period of the song collection, wherein the time period of the song collection is in inverse proportion to the preference of the first song;

Or, the song search parameter includes the time span of the song search, wherein the time span of the song search is in inverse proportion to the preference of the first song;

Or, the song playback parameter includes the song playback duration and/or the song playback duration, wherein the song playback duration is proportional to the preference of the first song, and the song playback duration is proportional to The preference of the first song is in an inverse proportional relationship.

Specifically, for the time length of the song collection, it refers to the time length from the time point of the song collection to the current time point. For example, the collection time point of the song A is 10 a.m today, and the song A is performed. The calculation time point of preference is today’s 8p.m. At this time, the collection of song A’s song collection is 10 hours (at this time, the collection of song A is in hours). The time point is January 1st, and the calculation time point for the preference degree of Song A is January 2nd. At this time, the time span of the song collection of Song A is 1 day (at this time, the time span is less than The number of days is in units). Wherein, if the number of times of song collection is multiple, then the time point of the song collection is the time point of the most recent song collection. It can be seen that the longer the time of the song collection is, the lower the user's preference for the song, and the preference for the song needs to be lowered.

For the time period since the song search, it refers to the time period from the time point of the song search to the current time point. For example, the search time point of song A is 10 a.m today, and the preference of song A is performed The calculation time point of is today’s 8p.m. At this time, the search time of song A’s song is 10 hours (at this time, the time since the present is in hours), if the search time of song A is On January 1st, and the calculation time point for the preference degree of Song A is January 2nd, at this time, the time span of the song search for Song A is 1 day (at this time, the time span from the present is in days as the unit ). Wherein, if the number of times of song search is multiple, then the time point of the song search is the time point of the most recent song search. It can be seen that the longer the song search time is, the longer the user's preference for the song is reduced, and at this time, the preference for the song needs to be lowered.

For the playing time of the song, it can refer to the total playing time of the song, or the playing time of the song at a certain time. For example, the i-th song A continues to be played for 3 minutes, then the i-th song A has a playing time of 3. minute. Generally, the longer the playing time of the song is, the higher the user's preference for the song is. At this time, the preference for the song needs to be increased.

For the duration of the song playback, it refers to the duration between the song playback time point and the current time point. For example, under the premise that all other conditions are the same, song A was played 1 day ago, and Song B was played 10 days ago. At this time, the preference of song A should be greater than the preference of song B. It can be seen that the longer the time since the song is played, the lower the user's preference for the song. At this time, the preference for the song needs to be lowered.

Therefore, the song collection parameters, song search parameters, and/or song playback parameters used above can be used to more accurately calculate the song recommendation results that meet the user's favorite needs.

Further as a preferred embodiment of the present invention, it also includes the following steps:

According to the priority corresponding to the song collection parameter, the priority corresponding to the song search parameter, the priority corresponding to the song playback parameter, and the priority corresponding to the in-vehicle perception information, output the song collection parameters and the priority The song search parameter, the song playback parameter and/or the song recommendation reason information corresponding to the vehicle-mounted perception information. The priority refers to the output priority of the song recommendation reason information corresponding to the parameter or information.

Specifically, if the preference of the currently recommended song relates to vehicle perception information, that is, the preference of the song has been numerically adjusted due to the vehicle perception information, and the song recommendation reason information corresponding to the vehicle perception information is output at this time: Recommend for you according to the current vehicle perception information; if the preference of the currently recommended song involves song collection parameters, the reason for outputting the song recommendation corresponding to the song collection parameter at this time is: According to your favorite single "X" And recommendation; if the preference of the currently recommended song involves song search parameters, the reason for outputting the song recommendation corresponding to the song search parameter at this time is: recommend based on the single "X" you searched for; if the currently recommended song The preference degree of is related to the song playing parameters. At this time, the reason for outputting the song recommendation corresponding to the song playing parameters is: recommend according to the single "X" you have heard. If the preference of the currently recommended song does not involve song collection parameters, song search parameters, song playback parameters, and vehicle perception information, the reason for the song recommendation output at this time is: recommendation according to your preferences. Among them, the output priority of the song recommendation reason information corresponding to the vehicle perception information>the output priority of the song recommendation reason corresponding to the song collection parameter>the output priority of the song recommendation reason corresponding to the song search parameter>the song The output priority of the song recommendation reason corresponding to the playback parameter. Generally, the preference for the first song involves song collection parameters, song search parameters, song playback parameters, and/or the vehicle perception information, while the preference for the second song involves vehicle perception information.

In addition, if the song recommendation reason information is output as the playlist recommendation reason information, that is, the recommendation is currently performed in the form of playlist recommendation, then for the playlist recommendation reason information, it is preferable to start from the playlist. The song with the highest preference is selected, and then output according to the above-mentioned song recommendation reason information output rule according to the preference of the selected song.

As shown in FIG. 2, the embodiment of the present invention also provides a song recommendation method, which preferably includes the following steps.

S201. Construct a song database and/or a playlist database.

Specifically, for the constructed song database and/or playlist database, it is mainly used to store the play volume, collection volume, release time, update time, tags and other attribute information of the song and/or playlist, and these attribute information Data can be provided for recommendation of songs and/or playlists.

S202: Acquire user behavior parameters of the first song. In this embodiment, the user behavior parameters include song collection parameters, song search parameters, and/or song playback parameters.

S203: Calculate the preference of the first song according to the user behavior parameter.

In this embodiment, the song collection parameter includes the time period of the song collection, wherein the time period of the song collection is inversely proportional to the preference of the first song; the song search parameter includes song search The time elapsed since the song search is inversely proportional to the preference of the first song; the song playback parameters include the time elapsed since the song is played and/or the time elapsed when the song is played. The playing time of the song is in a positive proportional relationship with the preference of the first song, and the playing time of the song is in an inverse proportion to the preference of the first song.

In order to calculate the preference of the first song based on the user behavior parameters more accurately, the following first formula is preferably used for calculation in this embodiment:

In the above formulas, X _'a represents the degree of preference for the song in a first header; _j1 the BT is expressed as a head of a first length in the first j1 song plays, that is, of a first section of the first song j1 The playback duration recorded in the second playback record is in minutes. It can be seen that BT _j1 belongs to the parameter of the song playback duration, and j1=1, 2, 3,..., B, and B is the first one. The total number of playback records of the song; BD _{j1 is} expressed as the length of time between the j1th playback time point of the ath first song and the current time point, which is in days. It can be seen that BD _j1 belongs to the distance of song playback The current time parameter; R indicates whether the user has collected this song (the first song of the a), if there is one, it is 1, otherwise, it is 0; RP indicates the weight of the song collection behavior, in this embodiment, The value of RP is 20 minutes, so it can be understood that under the same time attenuation, the collection behavior is equivalent to the playing duration of BT _j1 20 minutes; RD represents the time since the user’s song collection behavior occurred, that is, the distance between the song collection The current time is in days. Further, if there is only one song collection record in the collection behavior of the a-th first song, then the time since the collection record occurred is taken as the value of RD, If there are at least 2 song collection records in the collection behavior of the first song, then the time of the most recent song collection record should be taken as the value of RD; S indicates whether the user has searched for this Song (the first song of the ath), if there is one, it is 1, otherwise, it is 0; SP represents the weight of the song search behavior, in this embodiment, the value of SP is 10 minutes, so it can be understood as the same Under the time attenuation of, the search behavior is equivalent to the playing time of BT _j1 10 minutes; SD _j2 represents the time since the user performed the j2th search behavior on the first song a, that is, SD _j2 belongs to the song search distance The current time, where j2=1, 2, 3,..., S1, S1 is the total number of searches for the a-th first song; T1 and T2 are both time attenuation coefficients, they can be the same or different, and In this embodiment, they are the same and both take a value of 1/3, so it can be understood that the weight does not decay on the day the user behavior occurs, the weight decays to 1/2 of the original on the 8th day of the user behavior, and the weight on the 27th day The attenuation is 1/3 of the original, and the weight at the 81st day is attenuated to 1/4 of the original.

Or, for the preference of the first song, it may preferably be calculated using the following second formula:

The meanings of the symbols in the second formula are the same as those in the first formula.

S204. Calculate the user’s preference for different song characteristics according to the preference of the first song, and calculate the second song according to the user’s preference for different song characteristics and the weight of different song characteristics on the second song. 2. The preference of the song.

Specifically, in this embodiment, the step S204 includes:

S2041, placing preferences of different users for different first songs into a song preference matrix, where the number of rows in the song preference matrix represents the number of users, and the number of columns in the song preference matrix represents the number of songs;

S2042. After matrix decomposition is performed on the song preference matrix, the preference matrix for different song characteristics by different users and the weight matrix of different song characteristics for different songs are obtained, wherein the weight matrix of different song characteristics for different songs is Contains the weights of different song characteristics to different second songs;

S2043: After multiplying the favorite degree matrix and the weight matrix to obtain a song score matrix;

S2044. Obtain the preference degree of the second song from the song score matrix.

S205. After summing the preference degrees of several first songs included in the playlist and/or the preference degrees of several second songs included in the playlist, a recommendation degree of the playlist is obtained.

Specifically, in this embodiment, the recommendation of the playlist is preferably calculated using the following third calculation formula:

Or, use the following fourth calculation formula to calculate:

In the formula, Y _j3 is the recommendation degree of the j3th playlist; X _j4 is the preference degree of the j4th song in the j3th playlist. Among them, if the j4th song is the first song, then the steps The preference calculated in S203, if the j4th song is the second song, then the corresponding preference is obtained from the song score matrix in step S2044, j4=1, 2, 3,..., c, c Expressed as the total number of songs contained in the _{j3th playlist} ; L _j3 represents the audition volume of the j3th _playlist, that is, the popularity of the playlist; E represents the vehicle perception factor; G _j3 represents the number of hits of the interest tag; T _j3 It is expressed as the time since the j3th playlist update, that is, the time from the update time point to the current time point; T3 is expressed as the time attenuation coefficient, which may be the same as or different from T1 or T2, and in this embodiment , T3 is the same as T1.

Based on the above formula, after finding the recommendation degree of the playlist based on the preference of the first song and/or the preference of the second song, the songlist’s popularity, the updated duration of the playlist, and / Or the value of the number of hits of the interest tag to adjust the recommendation degree of the playlist, wherein the popularity of the playlist is in a proportional relationship with the recommendation degree of the playlist, and the time since the update of the playlist It is in an inverse proportional relationship with the recommendation degree of the playlist, and the number of hits of the interest tag is in a positive proportional relationship with the recommendation degree of the playlist, so that the recommended songs are not only more in line with the user’s taste, but also recommended updates are more popular The songs are given to users to improve the user experience of music listening.

For the vehicle perception factor E in the above formula, it is also used for real-time vehicle perception information (temperature, special weather and/or vehicle speed, etc.) obtained through vehicle perception, so as to adjust the recommendation degree of the song list, and the specific adjustment The methods include: 1. When the on-board sensor detects that the temperature parameter inside and/or outside the car is higher than the preset temperature threshold, the song list label has "burn", "hot", "fire" and other labels in the song list. The recommendation degree will have an extra on-board perception factor E added, that is, E is a non-zero positive number at this time; 2. When the vehicle speed is continuously lower than the preset speed threshold for a preset time period, it means that the vehicle continues to drive at a low speed. The recommendation degree of playlists with "light music", "quiet" and other tags in the single label will have an extra car perception factor E added, that is, E is a non-zero positive number at this time. That is to say, after obtaining the recommendation degree of the playlist according to the preference of the first song and/or the preference of the second song, the vehicle perception information will be acquired, and then the playlist tag matching the vehicle perception information will be obtained When it is judged that the playlist has the playlist tag, use E to increase the recommendation degree of the playlist, which not only enables the recommended playlist to meet the user’s taste, but also through recommended Songs to remind users of the current car environment, which is especially suitable for car scenes.

From the foregoing, the preference calculation in the embodiment of the present invention takes into account the user’s song playing, favorites, search behavior, and user interest factors (ie song characteristics), and the song list recommendation calculation also takes into account the popularity and update of the song list. Influencing factors such as time and vehicle factors, and the time factor T is also used to attenuate the effective degree of behavior, and the time attenuation of different behaviors can be different, and it also considers the behavior-based recommendation and the interest-based recommendation. However, the recommendation output result of the embodiment of the present invention is biased to prefer the behavior-based recommendation result. In addition, the preference algorithm of the embodiment of the present invention can support behavior superposition. When the behavior superimposes, the preference degree of songs and/or the recommendation degree of the playlist will be strengthened. For example, when a playlist hits both the playing behavior and the interest tag, The degree of recommendation will accelerate the enlargement; and also the popularity and update time of the playlist. Under the same behavioral preference, the hotter the playlist, the higher the recommendation, and the newer the playlist, the higher the recommendation.

S206. According to the priority corresponding to the song collection parameter, the priority corresponding to the song search parameter, the priority corresponding to the song playback parameter, and the priority corresponding to the vehicle-mounted perception information, output the song collection parameter , The song search parameter, the song playing parameter and/or the song recommendation reason information corresponding to the vehicle-mounted perception information.

Specifically, for the song recommendation reason information of any playlist, it is based on the song collection parameters, the song search parameters, the song playback parameters, and the currently acquired vehicle perception information related to the song with the highest preference in the playlist. The corresponding output is performed according to the recommended configuration rules, where the recommended configuration rules include:

Priority 1: When the preference calculation of the most preferred song in the playlist or the recommendation calculation of the playlist contains E and it is a non-zero positive number, then the song recommendation reason information for the playlist at this time is: Recommend to you based on current vehicle perception information;

Priority 2: When the song collection parameter is stored in the preference calculation of the song with the highest preference in the playlist, then the song recommendation reason information for the song list at this time is: Recommend according to your favorite single "X";

Priority 3: When the song search parameter is stored in the preference calculation of the song with the highest preference in the playlist, then the song recommendation reason information for the song list at this time is: recommend based on the single "X" you search;

Priority 4: When the song playback parameters are stored in the preference calculation of the song with the highest preference in the playlist, then the reason information for the song recommendation of the playlist at this time is: Recommend based on the single "X" you have listened to ；

Priority 5: Others, that is, when the preference calculation of the song with the highest preference in the playlist does not contain song collection parameters, song search parameters, song playback parameters, and vehicle perception information, then the song recommendation of the playlist The reason information is: Recommend according to your preferences. Among them, the above five priority levels are arranged in descending order: the first priority, the second priority, the third priority, the fourth priority, and the fifth priority.

S207. Output song recommendation results according to the recommendation degree of the playlist.

Specifically, among the recommended playlists, the recommendation degree of each playlist is calculated, and then according to the value of the playlist recommendation degree, the playlists are sorted from large to small, and then the sorted songs are sorted. After selecting several playlists that meet the recommendation rules, output the selected playlists to achieve song recommendation. For example, if you need to recommend 20 playlists at a time, then according to the number of playlist recommendation, from large to small Sort several playlists, and then select 20 playlists from the sorted playlists and output them. Among them, for the recommended rules, they include:

Rule 1: From high to low, the first priority, the second priority, the third priority and the fourth priority playlists are selected based on the recommended reasons. If the number of individual songs selected is less than 20, that is, the number of selected songs does not meet the required number, then select from the playlist with the fifth priority as the reason for recommendation, according to the recommendation degree of the playlist, from large to small. In order to make the final single number of selected songs 20.

Rule 2: In the same batch (20) of the selected playlist results, the recommended reasons for all playlists cannot be exactly the same, except for the reason for the fifth priority.

Rule 3: In the same batch (20) selected playlist results, the playlist cannot be repeated.

Rule 4: Recommended playlists that have been exposed in the past month cannot be recommended again; and recommended playlists that have not been exposed in the past month can be recommended again. And for this rule 4, if it is recommended for songs, then it is: according to the preference of the songs, sort the songs from largest to smallest, where the songs include the second song and the first song, and here In an embodiment, the first song requires that the generation time of its song playback, search and/or collection behavior has exceeded a preset time period (such as 1 month), otherwise, the first song is not recommended by the song. Select the range.

It can be seen from the above that, compared with traditional collaborative filtering, purely regular stack recommendation scoring mechanism sorting, etc., the embodiment of the present invention can greatly improve the fit between the final output recommendation result and the user’s favorite song, and greatly reduce The algorithm error (RMSE); at the same time, there is no need to establish a similarity relationship matrix between songs or playlists, which greatly improves computing efficiency. Moreover, different from the user interaction scenes of other music clients, the user experience scene is on the vehicle side, due to the particularity of its use scene (users are inconvenient to interact with the music client frequently, but the demand for listening to songs while driving is higher. ), therefore, the recommendation scheme implemented by the embodiment of the present invention is particularly suitable for application in vehicle-mounted scenarios. In particular, the calculation of preference and/or recommendation degree in the embodiment of the present invention also adds the influencing factor of vehicle-mounted perception information. The result of song recommendation is very suitable for in-vehicle scenarios.

As shown in Figure 3, an embodiment of the present invention also provides a song recommendation system, including:

The first obtaining unit is configured to obtain attribute parameters of the first song; the attribute parameters include at least one of song collection parameters, song search parameters, and song playback parameters;

The first calculation unit is configured to calculate the preference degree of the first song according to the attribute parameter;

The second calculation unit is used to calculate the user’s preference for different song characteristics according to the preference of the first song, and according to the user’s preference for different song characteristics and the preference of different song characteristics for the second song Weight, calculate the preference of the second song;

The recommendation unit is configured to output song recommendation results according to the preference degree of the first song and/or the preference degree of the second song.

It can be seen that the embodiment of the present invention uses the song collection behavior parameter, the song search behavior parameter and/or the song playback behavior parameter of the first song to calculate the preference of the first song, and then the preference of the first song is used to calculate The user’s preference for the song characteristics is calculated based on the user’s preference for the song characteristics and the weight of the song characteristics on the second song, so as to calculate the preference for the second song based on the preference of the first song and/or the second song It can be seen that the embodiment of the present invention combines the user's operating behaviors on the songs (including song collection behavior, search behavior, and playback behavior) to achieve the preference degree calculation of the first song, and uses the preference of the first song After calculating the user’s preference for the song characteristics, the second song’s preference is calculated. Therefore, compared with the traditional collaborative filtering and pure rule stacking recommendation, the result recommended by the embodiment of the present invention is It has a higher degree of fit with the user’s favorite song, that is, the recommendation result is more accurate, and there is no need to construct a similarity relationship matrix between songs or playlists in the calculation process of song recommendation. On the basis of the accuracy of the song recommendation, the processing efficiency can be greatly improved.

Further as a preferred embodiment of the system, the second calculation unit includes:

The first processing module is used to place different users’ preferences for different first songs into a song preference matrix, wherein the number of rows in the song preference matrix represents the number of users, and the number of columns in the song preference matrix Indicates the number of songs;

The second processing module is used to decompose the song preference matrix to obtain the preference matrix of different users for different song characteristics and the weight matrix of different song characteristics to different songs, wherein the different song characteristics have different pairs. The weight matrix of the song contains the weight of different song characteristics for different second songs;

The third processing module is configured to multiply the favorite degree matrix and the weight matrix to obtain a song score matrix;

The fourth processing module is used to obtain the preference degree of the second song from the song score matrix.

Further as a preferred embodiment of the system, the recommendation unit includes:

The first acquisition module is used to acquire vehicle-mounted perception information;

The second acquisition module is used to acquire a song tag matching the vehicle-mounted perception information;

The first adjustment module is used to increase the preference of the first song when it is determined that the first song has the song tag, and/or when it is determined that the second song has the song tag, Increase the preference of the second song;

The first recommendation module is configured to output song recommendation results according to the preference of the first song and/or the preference of the second song.

Or, the recommendation unit includes:

The first calculation module is used to sum up the preferences of several first songs included in the playlist and/or the preferences of several second songs included in the playlist to obtain the Recommendation

The second recommendation module is used to output song recommendation results according to the recommendation degree of the playlist.

Further as a preferred embodiment of the system, the second recommendation module includes:

The first acquisition sub-module is used to acquire vehicle sensing information;

The second acquisition sub-module is used to acquire a playlist tag matching the vehicle-mounted perception information;

The first adjustment submodule is configured to increase the recommendation degree of the playlist when it is determined that the playlist has the playlist tag;

The first recommendation submodule is used to output song recommendation results according to the recommendation degree of the playlist.

Further as a preferred embodiment of the system, it also includes:

The second acquiring unit is used to acquire the popularity of the playlist, the time since the update of the playlist, and/or the number of hits of interest tags;

The first adjustment unit is configured to adjust the recommendation degree of the playlist according to the popularity of the playlist, the time period since the update of the playlist, and/or the number of hits of the interest tag;

Wherein, the popularity of the playlist is in a positive relationship with the recommendation degree of the playlist, the time period since the update of the playlist is in an inverse proportional relationship with the recommendation degree of the playlist, and the number of hits of the interest tag is proportional to the song recommendation. The recommendation degree of a single is proportional.

Further as a preferred embodiment of the system, the song collection parameter includes the time period of the song collection, wherein the time period of the song collection is in inverse proportion to the preference of the first song;

Or, the song search parameter includes the time span of the song search, wherein the time span of the song search is in inverse proportion to the preference of the first song;

Or, the song playback parameter includes the song playback duration and/or the song playback duration, wherein the song playback duration is proportional to the preference of the first song, and the song playback duration is proportional to The preference of the first song is in an inverse proportional relationship.

Further as a preferred embodiment of the system, it also includes:

The reason output unit is configured to output the priority corresponding to the song collection parameter, the priority corresponding to the song search parameter, the priority corresponding to the song playback parameter, and the priority corresponding to the in-vehicle perception information. The song collection parameter, the song search parameter, the song playback parameter, and/or the song recommendation reason information corresponding to the vehicle-mounted perception information.

The contents of the above method embodiments are all applicable to this system embodiment, and the specific functions implemented by this system embodiment are the same as the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

As shown in FIG. 4, an embodiment of the present invention also provides a terminal, and the device includes:

At least one processor;

At least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the song recommendation method. Wherein, the terminal may be a client and/or server constituted by software and/or hardware.

It can be seen that the contents of the above method embodiments are all applicable to this terminal embodiment, and the specific functions implemented by this terminal embodiment are the same as the above method embodiments, and the beneficial effects achieved are the same as those achieved by the above method embodiments. the same. In addition, an embodiment of the present invention also provides a storage medium in which instructions executable by a processor are stored, and the instructions executable by the processor are used to execute the song recommendation method when executed by the processor. Similarly, the content in the above method embodiment is applicable to this storage medium embodiment, and the specific function implemented by this storage medium embodiment is the same as the above method embodiment, and the beneficial effects achieved are the same as those achieved by the above method embodiment. The beneficial effects are also the same.

The above is a detailed description of the preferred implementation of the present invention, but the invention is not limited to the described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. These equivalent modifications or replacements are all included in the scope defined by the claims of this application.

Claims (11)

1. A song recommendation method is characterized in that it comprises the following steps:

   Acquiring attribute parameters of the first song; the attribute parameters include at least one of song collection parameters, song search parameters, and song playback parameters;

   Calculate the preference of the first song according to the attribute parameter;

   According to the preference of the first song, the user’s preference for different song characteristics is calculated, and the second song is calculated according to the user’s preference for different song characteristics and the weight of different song characteristics on the second song Preference

   According to the preference degree of the first song and/or the preference degree of the second song, the song recommendation result is output.
2. The method for recommending songs according to claim 1, wherein the user’s preference for different song characteristics is calculated according to the preference of the first song, and according to the user’s preference for different song characteristics The degree value and the weight of different song characteristics on the second song, and the step of calculating the second song’s preference includes:

   Placing preferences of different users for different first songs into a song preference matrix, wherein the number of rows in the song preference matrix represents the number of users, and the number of columns in the song preference matrix represents the number of songs;

   After matrix decomposition of the song preference matrix, different users’ preference matrix for different song characteristics and a weight matrix of different song characteristics for different songs are obtained, wherein the weight matrix of different song characteristics for different songs includes The weight of different song characteristics to different second songs;

   After multiplying the favorite degree matrix and the weight matrix to obtain a song score matrix;

   The preference degree of the second song is obtained from the song score matrix.
3. The song recommendation method according to claim 1, wherein the step of outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song comprises:

   Obtain vehicle perception information;

   Acquiring a song tag matching the vehicle-mounted perception information;

   When it is determined that the first song has the song tag, increase the preference of the first song, and/or when it is determined that the second song has the song tag, increase the second Song preference;

   According to the preference degree of the first song and/or the preference degree of the second song, the song recommendation result is output.
4. The song recommendation method according to claim 1, wherein the step of outputting a song recommendation result according to the preference degree of the first song and/or the preference degree of the second song comprises:

   After summing the preferences of several first songs included in the playlist and/or the preferences of several second songs included in the playlist, the recommendation degree of the playlist is obtained;

   According to the recommendation degree of the playlist, the song recommendation result is output.
5. A song recommendation method according to claim 4, wherein the step of outputting the song recommendation result according to the recommendation degree of the playlist comprises:

   Obtain vehicle perception information;

   Acquiring a playlist tag matching the vehicle-mounted perception information;

   When it is determined that the playlist has the playlist tag, increase the recommendation degree of the playlist;

   According to the recommendation degree of the playlist, the song recommendation result is output.
6. A song recommendation method according to claim 4, characterized in that it further comprises the following steps:

   Get the popularity of the playlist, the time since the update of the playlist, and/or the number of interest tag hits

   Adjusting the recommendation degree of the playlist according to the popularity of the playlist, the time period since the update of the playlist, and/or the number of hits of the interest tag;

   Wherein, the popularity of the playlist is in a positive relationship with the recommendation degree of the playlist, the time period since the update of the playlist is in an inverse proportional relationship with the recommendation degree of the playlist, and the number of hits of the interest tag is proportional to the song recommendation. The recommendation degree of a single is proportional.
7. The method for recommending songs according to any one of claims 1-6, wherein the parameters of the song collection include the time length of the song collection, wherein the time length of the song collection is the same as that of the first song. The degree of preference is inversely proportional;

   Or, the song search parameter includes the time span of the song search, wherein the time span of the song search is in inverse proportion to the preference of the first song;

   Or, the song playback parameter includes the song playback duration and/or the song playback duration, wherein the song playback duration is proportional to the preference of the first song, and the song playback duration is proportional to The preference of the first song is in an inverse proportional relationship.
8. The method for recommending songs according to claim 5, further comprising the following steps:

   According to the priority corresponding to the song collection parameter, the priority corresponding to the song search parameter, the priority corresponding to the song playback parameter, and the priority corresponding to the in-vehicle perception information, output the song collection parameters and the priority The song search parameter, the song playback parameter and/or the song recommendation reason information corresponding to the vehicle-mounted perception information.
9. A song recommendation system is characterized in that it includes:

   The first obtaining unit is configured to obtain attribute parameters of the first song; the attribute parameters include at least one of song collection parameters, song search parameters, and song playback parameters;

   The first calculation unit is configured to calculate the preference degree of the first song according to the attribute parameter;

   The second calculation unit is used to calculate the user’s preference for different song characteristics according to the preference of the first song, and according to the user’s preference for different song characteristics and the preference of different song characteristics for the second song Weight, calculate the preference of the second song;

   The recommendation unit is configured to output song recommendation results according to the preference degree of the first song and/or the preference degree of the second song.
10. A terminal, characterized in that: the device includes:

    At least one processor;

    At least one memory for storing at least one program;

    When the at least one program is executed by the at least one processor, the at least one processor implements the song recommendation method according to any one of claims 1-8.
11. A storage medium storing instructions executable by a processor, wherein the instructions executable by the processor are used to execute a song according to any one of claims 1-8 when executed by the processor Recommended method.

Images