CN105787069A - Personalized music recommendation method - Google Patents

Personalized music recommendation method Download PDF

Info

Publication number
CN105787069A
CN105787069A CN201610116766.5A CN201610116766A CN105787069A CN 105787069 A CN105787069 A CN 105787069A CN 201610116766 A CN201610116766 A CN 201610116766A CN 105787069 A CN105787069 A CN 105787069A
Authority
CN
China
Prior art keywords
music
user
data
label
time
Prior art date
Application number
CN201610116766.5A
Other languages
Chinese (zh)
Inventor
刘海亮
徐倩倩
苏航
Original Assignee
中山大学深圳研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中山大学深圳研究院 filed Critical 中山大学深圳研究院
Priority to CN201610116766.5A priority Critical patent/CN105787069A/en
Publication of CN105787069A publication Critical patent/CN105787069A/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/60Information retrieval; Database structures therefor; File system structures therefor of audio data
    • G06F16/63Querying
    • G06F16/635Filtering based on additional data, e.g. user or group profiles

Abstract

The invention discloses a personalized music recommendation method. According to the method, different recommendations are made by judging the states of users, and for the simple recommendation, music which is located nearby and ranks the highest in number of listening times in a music library is directly selected; for the precise recommendation, the recommendation is divided into two parts, the latest music recommendation is conducted in the mode that a tag representing latest favor of the users is obtained through latest data, matching is conducted in the music library so as to obtain the corresponding latest music, while the nearby music recommendation is conducted in the mode that the most similar users are screened by way of secondary screening, and music recommendation is conducted through the users who are screened out. The whole process is that latest favor of the users is embodied, the nearby music data of the users are saved, when the music which the users does not like, a special formula is adopted for calculating playing times, and in order to reducing the calculated amount and avoid influence caused by artificial weight setting, the corresponding tag weight is calculated through data.

Description

The music of a kind of personalization recommends method

Technical field

The present invention relates to music and recommend field, particularly in solution cold start-up and recommending effectiveness, how solve emphatically by existing user's music data being analyzed and recommending best music to the problem of user.

Background technology

Along with the constantly universal of mobile equipment and development, user listens attentively to music anywhere or anytime by mobile equipment, and the music of the best how can be recommended to become a difficult problem to user.Existing recommendation method mainly has based on music content recommendation, based on musical correspondence recommendation, knowledge based recommendation, collaborative filtering recommending.

Recommending degree of accuracy not high for existing recommendation method, many recommendation method such as musical correspondence are recommended cannot give for personalized recommendation and cannot be considered by many factors when recommending simultaneously.

Patented method, considers user the operation of music, adopts mode record playback of songs number of times cleverly;Music is saved as by method two in part because the current hobby of user is ceaselessly changing;Take into full account the latest music attraction to user, increase latest music and recommend part;In life, environment is particularly significant on the impact of music, adopts and carries out music recommendation by the people around user, is taken in by environmental effect.

Summary of the invention

When analyzing existing recommendation method not enough, it is provided that a kind of music that degree of accuracy and personalized recommendation all take into full account recommendation method and system.

This invention address that the technical scheme that technical problem is taked is as follows

Judge whether user is new user or is not logged in user, if new user or be not logged in user and simply recommend, main obtain the music that near up-to-date music and user, people's great majority are liked and recommend

If user is old user, adopting the mode of accurate recommendation that user is carried out music recommendation, it mainly includes two parts, and respectively latest music is recommended and the recommendation of neighbouring music.

Latest music recommends mainly to comprise the steps:

A) music data that user listened for nearest n time is obtained, music data form (music ID, { label }, broadcasting time, which time).

B) the label weight in the music data of n time is calculated, owing to the broadcasting time in music data is to update with the ratio of the time span product of himself with this music number of times of broadcasting in this is play the time adopting this music to play in this is play, so this music is likely to do not liked by user when the broadcasting time of music is less than 1, so broadcasting time being not less than each music data of 1 calculate the label weight part as the weight of the label representing user of this music, and the incompatible hobby representing that user is current of the set of tags representing user of optimum is chosen by the threshold values set up.

C) calculating latest music numerical value, by the step b) tag combination obtained, the music deposited in music libraries also has respective label, corresponding numerical value is calculated by cosine similarity, being represented with 1 by the label of existence when calculating, non-existent label represents with 0, is calculated.

D) by numerical value, latest music is ranked up, calculates corresponding numerical value by step c), can be sorted accordingly by numerical value for up-to-date music

E) choose front m and first elect the latest music recommended

Neighbouring music recommends mainly to comprise the following steps that

A) all music datas listened of L user near user and this user are obtained, music data form (music ID, { label }, broadcasting time), music data for each user saves as two parts, one part is the music data listened for nearest n time, another part is all music datas listened, it is contemplated that user more recently by music for actual recommendation time purposes relatively big and historical data makes more greatly the more accurate of coupling when mating due to quantity of information.

B) the label weight of L user near this user and this user is calculated, the label weight of tag computation each of which label broadcasting time being not less than in 1 each music data, as the part of the weight of the label representing user, is multiplied by-1 simultaneously for broadcasting time less than the label weight of the corresponding label in the music data of 1.

C) calculate this user and the similar value of neighbouring L user, obtained the label weight representing user and nearby users by step b), after data are normalized, calculate corresponding similar value by cosine similarity.

D) calculate similar value by step c), by similar value, neighbouring L user is ranked up, K user before selecting.

E) music data that its front K the user selected in user and step d) listened for nearest n time is obtained.

F) calculate the label weight in the music data of nearest n time of user and neighbouring K user, the broadcasting time music data less than 1 is ignored and is not calculated.

G) calculate label weight by step f), calculated the corresponding numerical value of neighbouring K user by cosine similarity

H) by the step g) numerical value calculated, neighbouring K user is ranked up, chooses top n user

I) selected by step h) mate the most with user near N number of user, the music data of nearest n time of N number of user is carried out sort method, before obtaining, m is first-elected recommends music.

Accompanying drawing explanation

Accompanying drawing herein is merged in description and constitutes the part of this specification, explains we's ratio juris.

Fig. 1 is the overview flow chart of the inventive method.

Fig. 2 is latest music recommended flowsheet figure.

Fig. 3 is neighbouring music recommended flowsheet figure.

Detailed description of the invention

Specific embodiment of the invention step is as follows:

Fig. 1 lists the flow chart of a kind of friend recommendation method, as it is shown in figure 1, this friend recommendation method comprises the following steps S101-S104:

Step S101 is that the state to active user judges, if user is new user or the user that is not logged in adopts step S102, otherwise adopts step S103 to recommend.

Step S102 is the simple mode recommended, obtain the music data listened for nearest n time of M other users near user, the music data of neighbouring M other users is added up, by being ranked up selecting front X song to the music of statistics and the corresponding number of users listening this music, select front Y first equally by the number of users listened of the latest music deposited in music libraries and correspondence simultaneously, X and Y song for acquisition needs to carry out identical music processing only to retain a head, the music after acquisition process.

Step S103 is accurate recommendation mode, mainly includes latest music and recommends and the recommendation of neighbouring music, is determining that user has under the corresponding customized information of storage and completes, obtaining and recommend music accordingly.

The music that step S102 or S103 obtains is recommended corresponding user by step S104

Latest music in step S103 is recommended as in figure 2 it is shown, comprise the steps A1-A5:

Step A1 obtains the music data that user listened for nearest n time, music data form (music ID, { label }, broadcasting time, which time).Wherein label can have multiple, for the number of times C that music k i & lt is playkiComputing formula as follows:

(formula one)

Wherein t is the temporal summation that current music k i & lt was listened, TkIt is the time span of music k, QSecondaryIt it is this music k number of times play.Such as: data form corresponding to " white dreamer " be (10, { bass, guitar, national language, popular }, 2.5,1) representing that music ID in the server is 10, label includes { bass, guitar, national language, popular, broadcasting time is 2.5 times, listens when the last time listens music.

Step A2 calculates the label weight in the music data of nearest n time, chooses the label that can represent that user likes recently.First judge the broadcasting time of music, for number of times ignoring lower than 1, for instance: the broadcasting time of " white dreamer " correspondence is that 2.5 needs are calculated, and its label is { bass, guitar, national language, popular, represent that this user likes the bass of label recently, guitar, national language, popular label all increases by 0.625, and its computing formula is as follows:

(formula two)

Wherein ViBe music k label i & lt play in weight portion, CkiIt is the number of times of music k i & lt broadcasting, NNumber of tagsIt it is the number of tags of music k.

Each the label weight of all music that this user is nearest n timeThe label weight of all music is ranked up choosing the number that sets according to weight as the nearest hobby label of user from big to small.

Step A3 calculates the matching value of latest music, the hobby label nearest by the step A2 user chosen carries out cosine similarity calculating with all of latest music data in music libraries, wherein the label existed is represented with 1, it is absent from representing with 0, calculate the matching value between every first latest music and user, more close both the value of calculating is more little.

The step A4 similarity to being calculated by step A3 is ranked up according to order from small to large.

The sorted music of A4 is screened by step A5, m song before obtaining.

Neighbouring music in step S103 is recommended as it is shown on figure 3, comprise the steps B1-B9:

Step B1 obtains user and all history music datas of L other users individual, the form (music ID, { label }, broadcasting time) of these data near it.

Step B2 calculates the label weight of this user and neighbouring L user, for its labeling requirement of music less than 1 of broadcasting time, the value calculated is multiplied by-1 to ensure that the factor that user does not like is considered, and computing formula is formula two such as.

This user and the weighted value that near it, L user calculates are normalized respectively by step B3 by formula three, and the data after normalization are calculated this user and the similar value of neighbouring L user by cosine similarity.

y = x - x m i n x max - x m i n (formula three)

Wherein xminIt is need the minima in normalization data, xmaxIt is need the maximum in normalization data

The sequence from small to large of the similar value of neighbouring L the user that step B3 is calculated by step B4, chooses front K nearby users as the user similar to needing recommendation user.

Step B5 obtains user and the music data of nearest n time of K user near it.

Step B6 is the label weight in the music data of nearest n time that calculates user and neighbouring K user, for broadcasting time ignoring less than 1, is calculated by formula two.

Step B7 calculates this user and the matching value of neighbouring K user by cosine similarity, and its value participating in calculating is exactly the step B6 respective labels weight calculated.

Neighbouring K user is sorted from small to large by step B8 according to the step B7 matching value calculated, and chooses top n nearby users.

The music data that step B9 is nearest n time of N number of nearby users that step B8 is chosen is added up, identical music number of times is added, then sorting, choose and need to recommend user first at the front m not listened for nearest n time, the music identical for broadcasting time continues to sort with distance.

Content described in this specification embodiment is only enumerating of the way of realization to inventive concept; being not construed as of protection scope of the present invention is only limitted to the concrete form that embodiment is stated, protection scope of the present invention also and in those skilled in the art according to present inventive concept it is conceivable that equivalent technologies means.

Claims (7)

1. the method for a personalized recommendation, it is characterised in that including:
1) judge whether user is new user or is not logged in user.
2) if user is new user or is not logged in user, simply recommend, obtain the music data listened for nearest n time of M other users near user, the music data of neighbouring M other users is added up, by being ranked up selecting front X song to the music of statistics and the corresponding number of users listening this music, simultaneously first by Y before selecting after the number of users the listened sequence of the latest music deposited in music libraries and correspondence, X and Y song for acquisition needs to carry out identical music processing only to retain a head, the music after acquisition process.
3) if old user carries out accurate recommendation, mainly include latest music and recommend and the recommendation of neighbouring music, be determining that user has under the corresponding customized information of storage and completes, obtaining and recommend music accordingly.
4) latest music recommended characteristics is as follows:
A) music data that user listened for nearest n time is obtained, music data form (music ID, { label }, broadcasting time, which time).Wherein label can have multiple, for the number of times C that music k i & lt is playkiComputing formula as follows:
(formula one)
Wherein t is the temporal summation that current music k i & lt was listened, TkIt is the time span of music k, QSecondaryIt it is this music k number of times play.
B) calculate the label weight in the music data of nearest n time, choose the label that can represent that user likes recently.First judging the broadcasting time of music, for number of times ignoring lower than 1, corresponding computing formula is as follows:
(formula two)
Wherein ViBe music k label i & lt play in weight portion, CkiIt is the number of times of music k i & lt broadcasting, NNumber of tagsIt it is the number of tags of music k.
Each the label weight of all music that this user is nearest n timeThe label weight of all music is ranked up choosing the number that sets according to weight as the nearest hobby label of user from big to small.
C) matching value of latest music is calculated, the hobby label nearest by user carries out cosine similarity calculating with all latest music data in music libraries, wherein the label existed is represented with 1, be absent from representing with 0, calculate the matching value between every first latest music and user.
D) being ranked up according to matching value, before selecting, m head recommends.
5) near, music recommended characteristics is as follows:
A) user and the form (music ID, { label }, broadcasting time) of all history these data of music data of L other users individual near it are obtained.
B) calculating the label weight of this user and neighbouring L user, for its labeling requirement of music less than 1 of broadcasting time, the value calculated is multiplied by-1 to ensure that the factor that user does not like is considered, computing formula is formula two such as.
C) respectively this user and the weighted value that near it, L user calculates are normalized by formula three, the data after normalization are calculated by cosine similarity this user and the similar value of neighbouring L user.
y = x - x m i n x max - x m i n (formula three)
Wherein xminIt is need the minima in normalization data, xmaxIt is need the maximum in normalization data
D) to the sequence from small to large of the similar value of neighbouring L the user calculated, choose front K nearby users as to need the user that recommends user similar.
E) user and the music data of nearest n time of K user near it are obtained.
F) calculate the label weight in the music data of nearest n time of user and neighbouring K user, for broadcasting time ignoring less than 1, be calculated by formula two.
G) being calculated this user and the matching value of neighbouring K user by cosine similarity, its value participating in calculating is exactly the step B6 respective labels weight calculated, and is ranked up, and selects top n user.
H) music data of nearest n time of the N number of nearby users chosen is added up, identical music number of times is added, then sorting, choose and need to recommend user first at the front m not listened for nearest n time, the music identical for broadcasting time continues to sort with distance.
2. the music of a kind of personalization according to claim 1 recommends method, it is characterized in that: the music data of user is saved as two parts by this method, the music data of nearest n time is the embodiment that user likes recently, being the Primary Reference recommended, all of history music data is the comprehensive embodiment of user is the Primary Reference finding similar users.It is be calculated by formula one for the broadcasting time in data, is more refine tradition letter is unirecord, it is simple to find the music that user likes and dislikes.
3. the music of a kind of personalization according to claim 1 recommends method, it is characterized in that: step (2) considering, the recommendation of latest music and neighbouring music are recommended, fully user cannot be obtained for not knowing about of latest music the Resolving probiems of up-to-date music information, consider that the music that the most of users under similar environments like is recommended user by the impact of user by environment simultaneously.
4. the music of a kind of personalization according to claim 1 recommends method, it is characterized in that: the recommendation of music is divided into two parts by step (3), and unlike conventional recommendation general by music recommend simply mate, not only increase operand simultaneously for different music take unified mode cause coupling music be difficult to meet user.
5. the music of a kind of personalization according to claim 1 recommends method, it is characterised in that: step (4) adopts and calculates label weight by data and abandon tradition and think the unreasonable of definition weight.The computing formula of weight takes into full account that user ignores the broadcasting time music less than 1 for the broadcasting time of same song simultaneously because nearest data are likely to less employing, is calculated by computing formula two.Adopting there is tag representation when calculating matching value is 1 non-existent be expressed as 0 and be calculated by cosine similarity.
6. the music of a kind of personalization according to claim 1 recommends method, it is characterized in that: step (5) adopts all historical datas first passing through user carry out the screening to nearby users, carry out postsearch screening again through the music number number of up-to-date n time and obtain user the most similar.
7. the music of a kind of personalization according to claim 1 recommends method, it is characterized in that: step (5), is avoided when being normalized simultaneously for weight owing to historical data amount causes that label weight causes the excessive impact caused owing to the label of the relatively large music for broadcasting time less than 1 of the data volume of historical data participates in calculating as the label of the disagreeable music of user when calculating similar value.
CN201610116766.5A 2016-03-01 2016-03-01 Personalized music recommendation method CN105787069A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610116766.5A CN105787069A (en) 2016-03-01 2016-03-01 Personalized music recommendation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610116766.5A CN105787069A (en) 2016-03-01 2016-03-01 Personalized music recommendation method

Publications (1)

Publication Number Publication Date
CN105787069A true CN105787069A (en) 2016-07-20

Family

ID=56387585

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610116766.5A CN105787069A (en) 2016-03-01 2016-03-01 Personalized music recommendation method

Country Status (1)

Country Link
CN (1) CN105787069A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106446260A (en) * 2016-10-18 2017-02-22 刘洋 Self-learning music pushing method and system
CN106951068A (en) * 2017-02-23 2017-07-14 咪咕音乐有限公司 A kind of audio method for pushing and device
CN108062692A (en) * 2017-12-28 2018-05-22 平安科技(深圳)有限公司 Method, apparatus, equipment and computer readable storage medium are recommended in a kind of recording

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024058A (en) * 2010-12-31 2011-04-20 万音达有限公司 Music recommendation method and system
CN102654860A (en) * 2011-03-01 2012-09-05 北京彩云在线技术开发有限公司 Personalized music recommendation method and system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024058A (en) * 2010-12-31 2011-04-20 万音达有限公司 Music recommendation method and system
CN102654860A (en) * 2011-03-01 2012-09-05 北京彩云在线技术开发有限公司 Personalized music recommendation method and system

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106446260A (en) * 2016-10-18 2017-02-22 刘洋 Self-learning music pushing method and system
CN106951068A (en) * 2017-02-23 2017-07-14 咪咕音乐有限公司 A kind of audio method for pushing and device
CN108062692A (en) * 2017-12-28 2018-05-22 平安科技(深圳)有限公司 Method, apparatus, equipment and computer readable storage medium are recommended in a kind of recording

Similar Documents

Publication Publication Date Title
Celma Music recommendation
Sahoo et al. A hidden Markov model for collaborative filtering
CN102349072B (en) Identifying query aspects
Lin et al. Addressing cold-start in app recommendation: latent user models constructed from twitter followers
Wang et al. Opportunity model for e-commerce recommendation: right product; right time
Lamere Social tagging and music information retrieval
US20120209710A1 (en) System for Targeting Advertising to Mobile Communication Facilities Using Third Party Data
JP4433326B2 (en) Information processing apparatus and method, and program
US20090326947A1 (en) System and method for spoken topic or criterion recognition in digital media and contextual advertising
Hsu et al. AIMED-A personalized TV recommendation system
US20170364950A1 (en) Optimizing targeted advertisement distribution
Mandel et al. A web-based game for collecting music metadata
Aizenberg et al. Build your own music recommender by modeling internet radio streams
CN101641697B (en) Related search queries for a webpage and their applications
Quadrana et al. Sequence-aware recommender systems
Turnbull et al. A Game-Based Approach for Collecting Semantic Annotations of Music.
JP2009514075A (en) How to provide users with selected content items
EP2239670A1 (en) Information processing apparatus and method, and program thereof
US8060525B2 (en) Method and system for generating media recommendations in a distributed environment based on tagging play history information with location information
Romero et al. Personalized links recommendation based on data mining in adaptive educational hypermedia systems
US20080313146A1 (en) Content search service, finding content, and prefetching for thin client
CN102708130B (en) Calculate the easily extensible engine that fine point of user is mated for offer
US20080294655A1 (en) Method and apparatus for generation of a user profile
CA2924375A1 (en) Systems and methods for actively composing content for use in continuous social communication
Peltoniemi Cultural industries: Product–market characteristics, management challenges and industry dynamics

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20160720