CN106339505A - Music recommendation method based on Markov chain - Google Patents

Abstract

The invention belongs to the field of computer technology recommendation, and relates to a music recommendation method based on a Markov chain. The main method provided by the method comprises the following steps of: generating a scene assembly; establishing a matching matrix of the scene assembly and a recommendation algorithm library; and updating the matching matrix in the step b by adopting the Markov chain. The music recommendation method has the beneficial effects that users and music are analyzed from different scenes, the oneness of the recommendation method of the existing music recommendation system can be overcome, the matching matrix of the scenes and the algorithm, and the updating algorithm of the matching matrix of the scenes and the algorithm based on the Markov chain are provided, and the algorithm matching problem and the real-time music recommendation problem based on the scene under scene difference can be solved.

Description

A kind of music based on markov chain recommends method

Technical field

The invention belongs to computer technology recommends method field, it is related to a kind of music based on markov chain and recommends method.

Background technology

With the rapid electronic music market emerging, how to optimize and manage, and recommend suitable music to user It is the problem of an extremely important and rich challenge.Although the retrieval technique of music information is highly developed, music pushes away System of recommending but also rests on the primary stage.

In recent years, due to the emergence of Chinese online music platform and the popularization of smart machine, lead to China Electronics's music Market occurs in that long tail effect, and music itself also includes stronger emotion in addition.Therefore, traditional music recommendation algorithm has been The demand of user cannot be met.As improvement, the music recommendation algorithm currently using is mainly collaborative filtering and based on interior The proposed algorithm held.

But, find through actual investigation.Music and user's real-time scene present when listening to music have greatly Dependency.Different scenes is needed to recommend using different proposed algorithms, the variability issues of scene could be solved. Therefore, the matching problem of research scene and algorithm has real value.And because scene ceaselessly changes, algorithm is more New speed is that comparison is slow, solves to recommend problem to be extremely important and have real value based on the optimum in real time music of scene 's., as being described under current information, the stochastic process of following possible variation tendency, for update algorithm and scene for markov chain Coupling matrix be suitable.Therefore consider to solve the fraction of annual reporting law and scene matching matrix using the thought of markov chain more Newly, to adapt to the change of increasingly faster scene.Using the more New Policy of markov chain, algorithm can be made to keep up with the section of scene update Play, solve the problems, such as that optimum music is recommended in real time.

Content of the invention

The purpose of the present invention, it is simply that being directed to the problems referred to above, proposes a kind of music based on markov chain and recommends method.

The technical scheme is that a kind of music based on markov chain recommends method it is characterised in that including following walking Rapid:

A. generate scene set: the multiple users of statistics listen to scene during music, and enter rower respectively to different scenes Note, constitutes scene set using the multiple different scenes obtaining；

That b. sets up scene set and proposed algorithm storehouse mates matrix, specifically includes:

B1. described proposed algorithm storehouse is the set of existing music recommendation algorithm, then the expression formula of matrix m is as follows:

$m=\left\{\begin{array}{ccc}{x}_1{y}_1& \mathrm{...}& x_1{y}_n\\ .& .& .\\ .& .& .\\ .& .& .\\ x_a{y}_1& \mathrm{...}& x_a{y}_n\end{array}\right\}$

Wherein, x represents proposed algorithm, and y represents scene, and subscript n is numbered for scene, and subscript a is numbered for proposed algorithm, n and a It is the natural number more than or equal to 1；

B2. assume that in each scene, number of users is fixed, a proposed algorithm is calculated for one using equation below The score of the unique user in scene:

B3. the score according to unique user, calculates a proposed algorithm described in whole step b1 using equation below Score for whole scene:

$s_{{\mathrm{yx}}_1}=\frac{\underset{i=1}{\overset{n}{\σ}}{s}_i}{n}$

Wherein, n is total number of users under this scene；

B4. repeat step b2 and b3 are until obtain score under each scene for each algorithm, according to score and square Battle array m sets up scene set and the score in proposed algorithm storehouse, and to mate matrix s as follows:

$s=\left\{\begin{array}{ccc}{s}_{x_1{y}_1}& \mathrm{...}& s_{x_1{y}_n}\\ .& .& .\\ .& .& .\\ .& .& .\\ s_{x_a{y}_1}& \mathrm{...}& s_{x_a{y}_n}\end{array}\right\};$

C. using the method for markov chain, the coupling matrix obtaining in step b is updated, method particularly includes:

C1. user is carried out drawing section on a time period: using k user above as first time period section, by kth+1 to k+ M user as second time period, the like, often increase newly same user, then enter step c2 update a reuse algorithm Join matrix；

C2. the user to each section is it is assumed that the coupling matrix that t obtains is s^t, from t to t+1 moment, user couple Change in scene preference, the probability changing adopts equation below to calculate:

Subscript i, j is the numbering of different scenes；

Using formula:

$s_{ij}^{t+1}=\underset{k}{\σ}{s}_{ik}^t{t}_{kj}$

That is to say

s^t+1=s^t*t

Obtain the coupling matrix in t+1 moment, terminate to update.

Matrix s1 after renewal is as follows:

$s1=\left\{\begin{array}{ccc}\underset{k}{\σ}{t}_{x_{1}k}{s}_{x_1{y}_1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{1}k}{s}_{x_1{y}_n}\\ .& .& .\\ .& .& .\\ .& .& .\\ \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_a{y}_1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_a{y}_n}\end{array}\right\}.$

Beneficial effects of the present invention are from different scenes, user and music to be analyzed, and overcome existing music to push away Recommend the unicity of the proposed algorithm of system it is proposed that the coupling matrix of scene and algorithm, and the scene based on markov chain and calculation The update algorithm of the coupling matrix of method, solves under scene difference, the matching problem of algorithm and being pushed away based on the real-time music of scene Recommend problem.

Brief description

Fig. 1 is the schematic flow sheet of the present invention.

Specific embodiment

Below in conjunction with the accompanying drawings, describe technical scheme in detail:

As shown in figure 1, the main flow of the present invention is:

A. generate scene set: the multiple users of statistics listen to scene during music, and enter rower respectively to different scenes Note, constitutes scene set using the multiple different scenes obtaining；

That b. sets up scene set and proposed algorithm storehouse mates matrix, specifically includes:

B1. described proposed algorithm storehouse is the set of existing music recommendation algorithm, then the expression formula of matrix m is as follows:

$m=\left\{\begin{array}{ccc}{x}_1{y}_1& \mathrm{...}& x_1{y}_n\\ .& .& .\\ .& .& .\\ .& .& .\\ x_a{y}_1& \mathrm{...}& x_a{y}_n\end{array}\right\}$

Wherein, x represents proposed algorithm, and y represents scene, and subscript n is numbered for scene, and subscript a is numbered for proposed algorithm, n and a It is the natural number more than or equal to 1；

B2. assume that in each scene, number of users is fixed, a proposed algorithm is calculated for one using equation below The score of the unique user in scene:

B3. the score according to unique user, calculates a proposed algorithm described in whole step b1 using equation below Score for whole scene:

$s_{{\mathrm{yx}}_1}=\frac{\underset{i=1}{\overset{n}{\σ}}{s}_i}{n}$

Wherein, n is total number of users under this scene；

B4. repeat step b2 and b3 are until obtain score under each scene for each algorithm, according to score and square Battle array m sets up scene set and the score in proposed algorithm storehouse, and to mate matrix s as follows:

$s=\left\{\begin{array}{ccc}{s}_{x_1{y}_1}& \mathrm{...}& s_{x_1{y}_n}\\ .& .& .\\ .& .& .\\ .& .& .\\ s_{x_a{y}_1}& \mathrm{...}& s_{x_a{y}_n}\end{array}\right\};$

Scene set mate with the score in proposed algorithm storehouse matrix s be also denoted as shown in table 1 below:

Table 1 scene set mates matrix s with the score in proposed algorithm storehouse

C. using the method for markov chain, the coupling matrix obtaining in step b is updated, method particularly includes:

C1. user is carried out drawing section on a time period: using k user above as first time period section, by kth+1 to k+ M user as second time period, the like, often increase newly same user, then enter step c2 update a reuse algorithm Join matrix；

C2. the user to each section is it is assumed that the coupling matrix that t obtains is s^t, from t to t+1 moment, user couple Change in scene preference, the probability changing adopts equation below to calculate:

Subscript i, j is the numbering of different scenes；

Transfer matrix t is represented by as shown in table 2 below:

Table 2 transfer matrix t

Using formula:

$s_{ij}^{t+1}=\underset{k}{\σ}{s}_{ik}^t{t}_{kj}$

That is to say

s^t+1=s^t*t

Obtain the coupling matrix in t+1 moment, terminate to update.

Matrix s1 after renewal is as follows:

$s1=\left\{\begin{array}{ccc}\underset{k}{\σ}{t}_{x_{1}k}{s}_{x_1{y}_1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{1}k}{s}_{x_1{y}_n}\\ .& .& .\\ .& .& .\\ .& .& .\\ \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_a{y}_1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_a{y}_n}\end{array}\right\}.$

Claims (1)

1. a kind of music based on markov chain recommends method it is characterised in that comprising the following steps:

A. generate scene set: the multiple users of statistics listen to scene during music, and different scenes is marked respectively, adopt Constitute scene set with the multiple different scenes obtaining；

B. set up the matrix that mates of scene set and proposed algorithm storehouse, described proposed algorithm storehouse is existing music recommendation algorithm Set, then the expression formula of matrix m is as follows:

$m=\left\{\begin{array}{ccc}{x}_1{y}_1& \mathrm{...}& x_1{y}_n\\ .& .& .\\ .& .& .\\ .& .& .\\ x_a{y}_1& \mathrm{...}& x_a{y}_n\end{array}\right\}$

Wherein, x represents proposed algorithm, and y represents scene, and subscript n is numbered for scene, and subscript a is numbered for proposed algorithm, n and a is Natural number more than or equal to 1；

B1. assume that in each scene, number of users is fixed, a proposed algorithm is calculated for a scene using equation below In unique user score:

B2. the score according to unique user, using equation below calculate whole step b1 described in a proposed algorithm for The score of whole scene:

$s_{{\mathrm{yx}}_1}=\frac{\underset{i=1}{\overset{n}{\σ}}{s}_i}{n}$

Wherein, n is total number of users under this scene；

B3. repeat step b1 and b2, until obtaining score under each scene for each algorithm, build according to score and matrix m It is as follows that position scape set and the score in proposed algorithm storehouse mate matrix s:

$s=\left\{\begin{array}{ccc}{s}_{x_1{y}_1}& \mathrm{...}& s_{x_1{y}_n}\\ .& .& .\\ .& .& .\\ .& .& .\\ s_{x_a{y}_1}& \mathrm{...}& s_{x_a{y}_n}\end{array}\right\};$

C. using the method for markov chain, the coupling matrix obtaining in step b is updated, method particularly includes:

C1. user is carried out drawing on a time period with section: using k user above as first time period section, by kth+1 to k+m User as second time period, the like, often increase newly same user, then enter step c2 update a scene algorithmic match square Battle array；

C2. the user to each section is it is assumed that the coupling matrix that t obtains is s^t, from t to the t+1 moment, user is for field Scape preference changes, and the probability changing adopts equation below to calculate:

Subscript i, j is the numbering of different scenes；

Using formula:

$s_{ij}^{t+1}=\underset{k}{\σ}{s}_{ik}^t{t}_{kj}$

That is to say

s^t+1=s^t*t

Obtain the coupling matrix in t+1 moment, terminate to update；

Matrix s1 after renewal is as follows:

$s1=\left\{\begin{array}{ccc}\underset{k}{\σ}{t}_{x_{1}k}{s}_{x_{1}1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{1}k}{s}_{x_{1}n}\\ .& .& .\\ .& .& .\\ .& .& .\\ \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_{a}1}& \mathrm{...}& \underset{k}{\σ}{t}_{x_{a}k}{s}_{x_{a}n}\end{array}\right\}.$