CN105956040B - Song popularity analysis method in music information network under the influence of social networks - Google Patents

Abstract

The invention relates to a method for analyzing the popularity of songs under the influence of social relations in a music information network. First, the information subnet and the network mode of the information subnet are obtained from the music information network, and the social relations related to the information subnet are obtained, and then the information subnet is obtained from the music information network. Calculate the average edge betweenness of different types of relationship edges in the information subnet, and then calculate the propagation factors of different types of relationship edges according to the average edge betweenness, and then design edge weights in the information subnet according to the social relationship. Select nodes to perform random walk among the type nodes, and finally combine the propagation factors and edge weights to randomly walk in the information subnet to obtain the ranking of the nodes' influence in the network, so as to obtain the popularity of songs. The method proposed by the present invention takes into account the influence of user groups on songs, so that the final popularity analysis is more reasonable.

Description

Analysis method of song popularity under the influence of social relations in music information network

technical field

The invention relates to the field of information retrieval, in particular to a method for analyzing the popularity of songs under the influence of social relations in a music information network.

Background technique

The music information network is actually composed of an information subnet and a social subnet, and the information subnet usually contains songs and genre nodes related to songs, such as singers, lyricists, genres, etc. The relationships that exist in the network are song- The relationship between being sung and singing between singers, the relationship between being written and writing lyrics between songs and lyricists, the relationship between being included and included between songs and genres, etc. The social subnet includes friend relationships and songs between users. -Played-to-play relationship between users, contained-to-included relationship between users-groups.

At present, the analysis methods of song popularity in heterogeneous information networks basically use comprehensive structural information and rich semantic information such as different types of objects and relationships in the information subnet related to songs to analyze the popularity. However, this kind of information subnet pays more attention to the static information such as singers, lyricists, etc. related to songs, so the popularity of songs analyzed by traditional methods in this kind of network only indicates that in this information subnet Universality does not really reflect the popularity of songs in the user group; after the introduction of social relationships related to songs, considering the network differences between information subnets and social subnets, for example, the former often pays attention to Because of the static information in the network, the latter is more concerned with the network topology, so it cannot directly use the traditional method to analyze the popularity of songs in the whole network of music information network.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for analyzing the popularity of songs under the influence of social relations in the music information network, so as to overcome the problems existing in the methods for analyzing the popularity of songs in the existing music information network.

In order to achieve the above object, the technical scheme of the present invention is: a method for analyzing the popularity of songs under the influence of social relations in a music information network, the steps include the following:

Step S1: Obtain a music information network, extract information subnets without the influence of social relationships and network modes of the information subnets from the music information network, and extract information related to the information subnets from the music information network. social relationships;

Step S2: according to the network mode of the information subnet, calculate the average edge betweenness of different types of relation edges in the information subnet, and calculate the propagation factor of different types of relation edges in the random walk process according to the average edge betweenness;

Step S3: According to the social relationship, calculate the edge weights of various types in the information subnet under the influence of the social relationship;

Step S4: Perform random walks in the information sub-network in combination with the propagation factor and various types of edge weights, and calculate the influence ranking of various types of nodes under the influence of social relations, thereby obtaining the popularity of the song.

In an embodiment of the present invention, in the step S1, in the music information network, after removing the social relationship between users and the playback relationship and the played relationship between the user and the song, the music information only related to the song is obtained. The information subnet G _{0 without the influence of social relations has four types of nodes in the network mode of the information subnet G 0 ,} namely: songs, singers, genres, and lyricists.

In an embodiment of the present invention, the four types of nodes in the information sub-network have the following type relationships: the relationship between singer-song singing and being sung, lyricist-song relationship between writing lyrics and being written, genre- The inclusion and inclusion relationship between songs; and the user-song playing relationship is extracted from the music information network as the social relationship SG ₀ related to the information sub-network.

In an embodiment of the present invention, in the step S2, based on the network mode of the information subnet, the propagation factors of the following six types of edges are calculated, that is, the relationship between singing, being sung, writing lyrics, writing lyrics, including Relationships and included relationships are calculated as follows:

Step S21: acquiring the information subnet G ₀ ;

Step S22: Calculate the edge betweenness of the sung relationship edge, the lyric relationship edge, and the included relationship edge in the information subnet G ₀ , and calculate the average edge betweenness of the three types of edges The ratio of e _ms :e _ma :e _mt ;

Step S23: Calculate the propagation factors λ _ms , λ _ma , λ _mt of the three types of edges according to λ _ms :λ _ma :λ _mt =e _ms :e _ma :e _mt and λ _ms +λ _ma +λ _mt =1, And the propagation factors of the singing relationship edge, the lyric writing relationship edge, and the inclusion relationship edge are all 1, that is, λ _sm =1, λ _am =1, λ _tm =1.

In an embodiment of the present invention, in the step S3, the calculation process of each type of edge weight is as follows:

Step S31: The weight of the included relationship edge between song-genre is calculated as follows:

Among them, N _m ^T represents the number of genres to which the song m belongs, N _m ^U represents the number of users who play the song m, and P(t) represents the song list contained in the genre t;

Step S32: The weight of the inclusion relationship edge between genres and songs is calculated as follows:

Among them, N ^M _t represents the number of songs contained in genre t;

Step S33: The weight of the sung relationship edge between the song and the singer is calculated as follows:

Among them, N _m ^S represents the number of singers of song m, N _m ^U represents the number of users who play song m, and P(s) represents the list of songs sung by singer s;

Step S34: The weight of the singing relationship edge between singers and songs is calculated as follows:

in, Represents the number of songs sung by singer s;

Step S35: The weight of the lyricist relationship edge between the song and the lyricist is calculated as follows:

Among them, N _m ^A represents the number of lyricists of song m, N _m ^U represents the number of users who play song m, and P(a) represents the list of songs written by lyricist a;

Step S36: The weight of the lyric relationship edge between the lyricist and the song is calculated as follows:

Among them, N _a ^M represents the number of songs written by lyricist a.

In an embodiment of the present invention, the step S4 further includes the following steps:

Step S41: the edge weight W _mm between songs and songs in the information subnet, the edge weight W _ss between singers and singers, the edge weight W _aa between lyricists and the lyricists, and the edge weights between genres and genres. W _tt , singer-singer edge weight W _sa , lyricist-singer edge weight W _as , singer-genre edge weight W _st , genre-singer edge weight W _ts , lyricist-genre The edge weight W _at and the genre-writer edge weight W _ta are all set to 0, namely W _mm ,W _ss ,W _aa ,W _tt ,W _sa ,W _as ,W _st ,W _ts ,W _at ,W _ta are set to zero matrices of corresponding sizes; the spread factor between song-song λ _mm , the spread factor between singer-singer λ _ss , the spread factor between lyricist and lyricist λ _aa , the spread factor between genres and genres λ _tt , singer-singer propagation factor λ _sa , lyricist-singer propagation factor λ _as , singer-genre propagation factor λ _st , genre-singer propagation factor λ _ts , lyricist-genre propagation factor λ ts The inter-distribution factor λ _at and the genre-writer propagation factor λ _ta are all set to 0, that is, λ _mm = λ _ss = λ _aa = λ _tt = λ _sa = λ _as = λ _st = λ _ts = λ _at = λ _ta = 0;

Step S42: In the information subnet, the transition probability matrix TPM for random walk between nodes is calculated as follows:

Step S43: Set two vectors of length N, Vec_C and Vec_R; the value in Vec_R is the influence value of each node in the information subnet, and Vec_C is initially and calculate Vec_R as follows:

Among them, ε is the random jump probability of the whole graph, and N is the total number of nodes in the information subnet;

Step S44: Calculate Vec_C and Vec_R by the following two formulas:

Vec_C=Vec_R,

And when ||Vec_R-Vec_C||≥ξ, continue the calculation of the above two formulas, otherwise get Vec_R, and ξ is the preset error threshold;

Step S45: Sort the genre node songs, singers, lyricists and genre values in Vec_R respectively to obtain the sorted result sequence of songs, singers, lyricists and genres, and then obtain the popularity of the songs.

Compared with the prior art, the present invention has the following beneficial effects: the present invention proposes a method for analyzing the popularity of songs under the influence of social relations in a music information network, which not only effectively utilizes the different information in the information sub-network related to songs The comprehensive structural information and rich semantic information such as types of objects and relationships, while taking into account the influence of user groups on songs, can avoid the difference between the information subnet and the social subnet in the music information network. This method is more reasonable to analyze the popularity of songs with the help of the social relations of the social subnet in the information subnet.

Description of drawings

FIG. 1 is a method for analyzing the popularity of songs under the influence of social relations in a music information network according to the present invention.

FIG. 2 is a music information network in an embodiment of the present invention.

FIG. 3 is a network mode of an information subnet in a music information network according to an embodiment of the present invention.

FIG. 4 is an example of an information subnet under the influence of social relations in a music information network in an embodiment of the present invention.

Detailed ways

The technical solutions of the present invention will be described in detail below with reference to the accompanying drawings.

The present invention will be further described below by specific embodiments, but it should be noted that the purpose of publishing the embodiments is to help further understand the present invention, but those skilled in the art can understand: without departing from the present invention and the appended claims Various substitutions and modifications are possible within the spirit and scope of the Therefore, the present invention should not be limited to the contents disclosed in the embodiments, and the scope of protection of the present invention shall be subject to the scope defined by the claims.

As shown in Figure 1, it is a flowchart of a method for analyzing the popularity of songs under the influence of social relationships in a music information network proposed by the present invention, and the method includes the following steps:

Step S1 : acquiring the music information network, extracting the information subnet without the influence of social relations and the network mode of the information subnet from the music information network, and extracting the social relations related to the information subnet from the music information network.

Further, in this embodiment, in the music information network, after removing the social information such as the social relationship between users and the playing relationship and the played relationship between the user and the song, it is obtained only related to the song without the influence of the social relationship. Below the information subnet G ₀ , in the network mode of this information subnet G ₀ , there are four types of nodes, namely songs, singers, genres, and lyricists. The relationships in the network are: singer-song The relationship between singing and being sung, the relationship between lyricist and lyricist between the lyricist and the song, and the relationship between genre-song and inclusion; and then extract the playback relationship between users and songs from the music information network, that is, with Information subnet-related social relations SG ₀ .

Further, in this embodiment, music network data information is obtained from the Douban Music website, and network entities are extracted, including song (M), singer (S), lyricist (A), genre (T), user (U) , grouping (G), there are relationships in these data: sung and sung relationship between song-singer, lyric and lyric relationship between song-lyricist, included and included relationship between song-genre, The relationship of playing and being played between users and songs, and the relationship between grouping and containing and being included between users, so as to generate a music information network. An example of the music information network is shown in FIG. 2 .

Further, in this embodiment, there are node songs (M), singers (S), lyricists (A), genres (T) in the network mode of the information subnet in the music information network, and edges represent the relationship between nodes. , which are the relationship between being sung and singing between songs and singers, the relationship between being written and writing lyrics between songs and lyricists, and the relationship between being included and included between songs and genres. The network mode of the information subnet is shown in Figure 3. Show. The social relationship related to the information subnet refers to the relationship between user and song being played and played.

Step S2: Calculate the average edge betweenness of different types of relation edges in the information subnet according to the network mode of the information subnet, and calculate the propagation factors of different types of relation edges in the random walk process according to the average edge betweenness.

Further, based on the network mode of the information subnet, the propagation factors of the following six types of edges are calculated, namely, the singing relationship, the singing relationship, the lyric relationship, the lyric writing relationship, the inclusion relationship, and the included relationship. The calculation method is as follows:

Step S21. Obtain the information subnet G ₀ ;

Step S22. Calculate the edge betweenness of the sung relationship, the lyric relationship, and the included relationship edge in the information subnet G ₀ , and count the ratio of the average edge betweenness of these three types of edges _ems :e _ma :e _mt ;

Step S23. According to λ _ms : λ _ma : λ _mt =e _ms :e _ma :e _mt and λ _ms +λ _ma +λ _mt =1, calculate the propagation factors λ _ms , λ _ma , λ _mt of these three types of edges , and the propagation factors of the singing relation side, the lyric writing relation side and the inclusion relation side are all 1, that is, λ _sm =1, λ _am =1, λ _tm =1.

As shown in FIG. 4 , in this embodiment, e _ms :e _ma :e _mt =9.5332:7.5288:8.7495, λ _ms =0.3693, λ _ma =0.2917, λ _mt =0.3390.

Step S3: Designing various types of edge weights in the information subnet G ₀ under the influence of the social relationship based on the network mode of the information subnet and according to the social relationship, and the specific calculation formula is as follows:

Step S31: The weight calculation formula of the included relation edge between songs and genres: N _m ^T represents the number of genres to which the song m belongs, N _m ^U represents the number of users who play the song m, and P(t) represents the list of songs included in the genre t;

Step S32: The formula for calculating the weight of the inclusion relationship edge between genres and songs: N _t ^M represents the number of songs contained in genre t;

Step S33: The weight calculation formula of the sung relationship edge between the song and the singer: N _m ^S represents the number of singers of song m, N _m ^U represents the number of users who play song m, and P(s) represents the list of songs sung by singer s;

Step S34: The weight calculation formula of the singing relationship edge between singer-song: N _s ^M represents the number of songs sung by singer s;

Step S35: the calculation formula of the weight of the lyric relationship edge between the song and the lyricist: N _m ^A represents the number of lyricists of song m, N _m ^U represents the number of users who play song m, and P(a) represents the list of songs written by lyricist a;

Step S36: The weight calculation formula of the lyric relationship edge between the lyricist and the song: N _a ^M represents the number of songs written by lyricist a;

Step S4: Combine the propagation factor and the edge weight to perform a random walk in the information subnet, and obtain the influence ranking of the four types of nodes: song, singer, lyricist, and genre. The specific calculation method is as follows:

Step S41. Put the song-song edge weight W _mm in the network, the singer-singer edge weight W _ss , the lyricist-lyricist edge weight W _aa , the genre-genre edge weight W _tt , the singer- Edge weight between lyricists W _sa , edge weight between lyricist-singer W _as , edge weight between singer-genre W _st , edge weight between genre-singer W _ts , edge weight between lyricist-genre W _at , The edge weight W _ta between genres and lyricists is set to 0. In this embodiment, W _mm is a 6×6 zero matrix, W _ss is a 4×4 zero matrix, and W _aa is a 2×2 zero matrix Matrix, W _tt is a 3×3 zero matrix, W _sa is a 4×2 zero matrix, W _as is a 2×4 zero matrix, W _st is a 4×3 zero matrix, W _ts is a 3×4 zero matrix matrix, W _at is a 2×3 zero matrix, W _ta is a 3×2 zero matrix; the song-song propagation factor λ _mm , the singer-singer propagation factor λ _ss , the lyricist-lyricist Propagation factor λ _aa , genre-genre propagation factor λ _tt , singer-songwriter propagation factor λ _sa , lyricist-singer propagation factor λ _as , singer-genre propagation factor λ _st , genre-singer The propagation factor λ _ts between lyricists, the propagation factor λ _at between lyricists and genres, and the propagation factor λ _ta between genres and lyricists are all set to 0, that is, λ _mm = λ _ss = λ _aa = λ _tt = λ _sa = λ _as =λ _st =λ _ts =λ _at =λ _ta =0;

Step S42. In the information subnet, two points need to be considered when a node performs a random walk, that is, what type of node should be selected in the next random walk and which specific node in the set of nodes of this type should be selected. In fact, the former is a variety of The propagation factor of type edges, the latter is the weight of various types of edges, therefore, the transition probability from node i to node j is the product of these two, that is, the calculation formula of transition probability matrix TPM is as follows:

Step S43. Set two vectors of length N, Vec_C, Vec_R, and the values in Vec_R are the influence values of each node in the information subnet, and Vec_C is initially pass The formula calculates Vec_R; in this embodiment, N=15;

Step S44. Calculate Vec_C and Vec_R by the following two formulas

Vec_C=Vec_R

When ||Vec_R-Vec_C||≥ξ, continue the calculation of the above two formulas, otherwise Vec_R is obtained, and ξ is the pre-set error threshold;

Step S45. Sort songs, singers, lyricists and genre values in Vec_R respectively, obtain the sorting result sequence of the four types of nodes of song, singer, lyricist and genre, and obtain song popularity, as shown in Table 1.

Table 1

The above are the preferred embodiments of the present invention, all changes made according to the technical solutions of the present invention, when the resulting functional effects do not exceed the scope of the technical solutions of the present invention, belong to the protection scope of the present invention.

Claims (2)

1. A song popularity analysis method under the influence of social relations in a music information network is characterized by comprising the following steps: the method comprises the following steps:

step S1: acquiring a music information network, extracting an information subnet and a network mode of the information subnet without the influence of social relations from the music information network, and extracting the social relations related to the information subnet from the music information network;

step S2: calculating the average edge betweenness of different types of relation edges in the information subnet according to the network mode of the information subnet, and calculating the propagation factors of the different types of relation edges in the random walk process according to the average edge betweenness;

step S3: calculating various types of side weights in the information subnet under the influence of the social relationship according to the social relationship;

step S4: combining the propagation factors and various types of edge weights to carry out random walk in the information subnet, and calculating various types of node influence sequencing under the influence of social relations, thereby obtaining song popularity;

in step S1, after removing the social relationship between the users and the playing relationship and played relationship between the users and the songs in the music information network, an information subnet G related to the songs only and without the influence of the social relationship is obtained₀The information subnet G₀The network mode of (1) has four types of nodes, which are respectively: song, singer, genre, and word writer;

the following type relationships exist among four types of nodes in the information sub-network: the relation between singers and songs, the relation between word makers and songs, and the relation between inclusion and inclusion between genres and songs; and extracting the playing relation between the user and the song from the music information network as the social relation SG related to the information subnet₀；

In step S2, based on the network model of the information subnet, propagation factors of the following six types of edges, i.e., singing relation, word-making relation, inclusion relation, and inclusion relation, are calculated according to the following steps:

step S21: obtaining the information subnet G₀；

Step S22: calculating said information subnet G₀The edge betweenness of the sung relation edge, the word relation edge and the included relation edge, and calculating the ratio e of the average edge betweenness of the three types of edges_ms:e_ma:e_mt；

Step S23: according to λ_ms:λ_ma:λ_mt＝e_ms:e_ma:e_mtAnd λ_ms+λ_ma+λ_mtCalculating the propagation factor lambda of the three types of edges as 1_ms,λ_ma,λ_mtAnd the propagation factors of the singing relation edge, the word making relation edge and the inclusion relation edge are all 1, namely lambda_sm＝1,λ_am＝1,λ_tm＝1；

In step S3, the calculation procedure of each type edge weight is as follows:

step S31: the weights of the contained relationship edges between song-genres are calculated as follows:

wherein N is_m ^TIndicates the number of genres to which song m belongs, N_m ^URepresents the number of users playing song m, and p (t) represents a song list contained in genre t;

step S32: genre-weight of inclusion relation edges between songs is calculated as follows:

wherein N is_t ^MRepresents the number of songs included in genre t;

step S33: the weight of the sung relationship edge between song-singer is calculated as follows:

wherein N is_m ^SIndicates the number of singers of Song m, N_m ^UA list of songs representing the number of users playing song m, p(s) representing singers singing;

step S34: the weights of the singer-song singing relationship edges are calculated as follows:

wherein N is_s ^MRepresents the number of songs sung by singers;

step S35: the weights of the wordled relationship edges between song-worders are calculated as follows:

wherein N is_m ^ANumber of word makers of Song m, N_m ^UA song list representing the number of users playing song m, and p (a) representing the word writer a;

step S36: the weights of the wordwise relationship edges between the worders-songs are calculated as follows:

wherein N is_a ^MIndicating the number of songs the word writer a makes.

2. The method for analyzing popularity of songs under the influence of social relationships in a music information network as claimed in claim 1, wherein said step S4 further comprises the steps of:

step S41: weighting W between songs in the information subnet_mmThe side weight W between the singer and the singer_ssThe edge weight W between the word writer and the word writer_aaGenre-genre boundary weight W_ttInter-singer-word-maker side weight W_saSide weight W between speaker and singer_asInter-artist-genre weight W_stGenre-singer side weight W_tsEdge weights W between speaker-genres_atSide weight W between genre and author_taAre all set to 0; transmitting a song-song propagation factor lambda_mmSinger-singer propagation factor lambda_ssPropagation factor lambda between word makers and word makers_aaGenre-genre spread factor lambda_ttPropagation factor lambda between singer and word writer_saPropagation factor lambda between speaker and singer_asSinger-genre transmission factor lambda_stGenre-singer propagation factor lambda_tsInterwriter-genre propagation factor lambda_atGenre-to-writer propagation factor lambda_taAre all set to 0, i.e. λ_mm＝λ_ss＝λ_aa＝λ_tt＝λ_sa＝λ_as＝λ_st＝λ_ts＝λ_at＝λ_ta＝0；

Step S42: in the information subnet, a transition probability matrix TPM for random walk among nodes is calculated as follows:

step S43: setting two vectors Vec _ C and Vec _ R with the length of N; the value in Vec _ R is the influence value of each node in the information subnet, and Vec _ C is initiallyAnd Vec _ R is calculated as follows:

wherein epsilon is the whole-graph random jump probability, and N is the total number of nodes in the information subnet;

step S44: vec _ C and Vec _ R are calculated by the following two formulas:

Vec_C＝Vec_R，

when | | | Vec _ R-Vec _ C | | | | is equal to or greater than ξ, continuing the calculation of the two formulas, otherwise, obtaining Vec _ R, wherein ξ is a preset error threshold;

step S45: and respectively sequencing the songs, singers, word makers and genre values of the type nodes in the Vec _ R to obtain sequencing result sequences of the songs, the singers, the word makers and the genres, and further obtain the song popularity.