CN106528653B - A kind of context-aware music recommended method based on figure incorporation model - Google Patents
A kind of context-aware music recommended method based on figure incorporation model Download PDFInfo
- Publication number
- CN106528653B CN106528653B CN201610902186.9A CN201610902186A CN106528653B CN 106528653 B CN106528653 B CN 106528653B CN 201610902186 A CN201610902186 A CN 201610902186A CN 106528653 B CN106528653 B CN 106528653B
- Authority
- CN
- China
- Prior art keywords
- music
- user
- vector
- interest
- context
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/63—Querying
- G06F16/635—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/686—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings
Abstract
The context-aware music recommended method based on figure incorporation model that the invention discloses a kind of, comprising: the extraction of musical features of the S1. based on figure incorporation model;S2. the extraction and modeling of user's overall situation interest in music and context interest in music;S3. the music of context-aware is recommended.The present invention extracts the feature of music using figure incorporation model from the metadata of the played data of user and music, the global interest in music and context interest in music for recording and playing acquisition user in record in the recent period are played from the complete of user again, the global interest and current context interest of user are finally comprehensively considered when recommending, so as to allow the music of recommendation to meet the real-time requirement and preference of user, to reduce the searching cost of user and improve the satisfaction of user.
Description
Technical field
The invention belongs to data mining and recommended technology fields, and in particular to a kind of context sense based on figure incorporation model
The happy recommended method of bosom friend.
Background technique
With the development of digital music industry, there are more and more online digital music providers, allows user can
To listen to the music liked whenever and wherever possible, such as apple Online Music shop is provided more than 30,000,000 first digital musics.Meanwhile
The music data of magnanimity increases the difficulty that user finds its interested music.In addition, user's listens song hobby would generally be with
The time, space, weather, physical condition is different and changes, traditional music recommender system has not been suitable for personalized mobile network
Network service field.In recent years, the mixing music recommender system based on context-aware is by by contextual information and auxiliary information
Recommender system is introduced, an emerging research field is become.It finds under study for action, contextual information and auxiliary information involvement is pushed away
System is recommended, is equivalent to and traditional " user-project " two dimension scoring utility models is extended to comprising a variety of contextual informations and auxiliary
The multidimensional scoring utility models of supplementary information, are conducive to improve and recommend accuracy.Therefore the mixing music based on context-aware pushes away
System is recommended with important research significance.
Currently, the music mix recommended method of context-aware, which generally employs " multidimensional recommendation ", is converted into " two dimension is recommended "
Mentality of designing filtered before recommendation results generation, after generating or during generate using current context information
Fall with the unmatched data of current context information, while using conventional two-dimensional recommended technology (comprising collaborative filtering, be based on content
Filtering, Knowledge based engineering filtering, composite filtering etc.) generate recommendation results.Because the maturation of conventional recommendation systems is utilized
Technology, such method become context-aware recommended method most widely used at present.
However, the prior art only considered the contextual information and portion of secondary of user in the matching process of music and user
It helps, lacks and the deep layer of music content is parsed, it is believed that all music are all homogeneities, and the different attribute of music is from user not
With the different fancy grades having under situation to music, i.e., differentiation differentiation is carried out to different music by the user property of music,
To have ignored music as a kind of multimedia file, its own context property having.This recommended method is excessively subjective,
The coupling of user and music is reduced, and is not bound with the auxiliary information of music, such as play sequence and the first number of music
According to so that the precision on recommender system is influenced.Under many scenes, the context of listening to of user often dominates user's
Demand, such as the global preferences of user include rock music and absolute music, but user at night rest when, after preferring
Person.
Summary of the invention
For above-mentioned technical problem present in the prior art, it is upper and lower based on figure incorporation model that the present invention provides a kind of
Text perception music recommended method, can allow the music of recommendation to meet the real-time requirement and preference of user.
A kind of context-aware music recommended method based on figure incorporation model, includes the following steps:
(1) entire music for collecting user listens to the label information of sequence and all music;The entire music is listened to
Sequence, which includes user's history, listens to record for every of music, and the label information includes singer's (or performance of music
Person), affiliated album and style of song type;
(2) metadata of sequence and all music is listened to according to the entire music of all users, establishing includes user-sound
Happy interaction figure, music-music transfer figure and music-label knowledge graph Heterogeneous Information model;
(3) according to the following objective function L of the Heterogeneous Information model foundation:
Wherein:Indicate music m in music-music transfer figureiWith music mjEdge-stitching weight,It indicates
Music miWith music mjIndirect relation weight, p (mi,mj) indicate music miWith music mjThe probability listened to together, M indicate by
The music storehouse of all music compositions;
(4) minimum solution is carried out to above-mentioned objective function L, in the hope of the feature vector of per song in music storehouse M;
(5) all listen in sequence is listened to user's entire music and records corresponding musical features vector averaging, obtained
The global interest in music vector of user;
(6) it is formed closely from the record of listening to interior for the previous period that user's entire music listens to extraction current time in sequence
Phase music listens to sequence;And then all listen in sequence is listened to recent music and records corresponding musical features vector averaging,
Obtain the context interest in music vector of user;
(7) according to the feature vector of per song and the global interest in music vector sum context interest in music of user to
Amount, calculates user for the interest value of per song;And then all music in music storehouse are arranged from big to small according to interest value
Sequence, and extract the maximum several songs of interest value and recommend user.
It is connected between user and music by sideline in the user-music interaction figure, wherein user u and music mjBetween
The weight in sidelinetU, mjMusic m is listened to for user ujNumber, u ∈ U, U are indicated by all user groups
At user's cluster.
It is connected between music and music by sideline in the music-music transfer figure, wherein music miWith music mjBetween
The weight in sidelinetMi, mjFor music miWith music mjThe number listened to together, i.e., same complete
If music listens to music m in sequenceiWith music mjCorresponding listen to is recorded within the scope of certain intervals, then determines music miAnd sound
Happy mjIt is listened to simultaneously.
It is connected between music and label by sideline in the music-label knowledge graph, wherein music mjBetween label d
The weight in sidelinetD, mjFor music mjIt is endowed the number of label d, d ∈ D, D expression is by all labels
The tag library of composition.
The indirect relation weightExpression formula it is as follows:
Wherein:For music m in user-music interaction figureiAll sideline weights composition vector,For with
Music m in family-music interaction figurejAll sideline weights composition vector,For music m in music-label knowledge graphi's
The vector of all sideline weight compositions,For in music-label knowledge graph music mj all sideline weights composition vector,
Cos (A, B) indicates the cosine similarity of vector A and vector B,Indicate the concatenation of two vectors.
Probability p (the mi,mj) expression formula it is as follows:
Wherein:WithRespectively music miWith music mjFeature vector,TIndicate vector transposition.
User is calculated by the following formula for the interest value of per song in the step (7):
Wherein: xu,iIt is user u for music miInterest value,WithRespectively music miWith music mjFeature to
Amount, vuFor the global interest in music vector of user u, zuFor the context interest in music vector of user u,TIndicate vector transposition, u ∈
U, U indicate the user's cluster being made of all users.
The present invention utilizes figure incorporation model to obtain the spy of music from the metadata of the played data of user and music for the first time
The global interest characteristics vector for levying vector sum user, provides a kind of reliable solution for music features extraction difficulty;This
Invention is emerging according to the global interest in music and context music of the feature vector acquisition user of the music in the broadcasting record of user
Interest provides a kind of feasible thinking for the difficult problem of the extraction and modeling of the interest of user;It is emerging to comprehensively consider user's overall situation
The recommended method of interest and context interest, the present invention enable to the music recommended more to meet the current preference of target user, from
And it reduces the searching cost of user and improves the satisfaction of user.
Detailed description of the invention
Fig. 1 is the system architecture schematic diagram of music recommended method of the present invention.
Fig. 2 is the flow diagram of user's music preferences prediction in music recommended method of the present invention.
Fig. 3 is to construct Information heterogeneity mould using the metadata of user's played data and music in music recommended method of the present invention
The schematic diagram of type.
Specific embodiment
In order to more specifically describe the present invention, with reference to the accompanying drawing and specific embodiment is to technical solution of the present invention
It is described in detail.
The present invention is based on the music recommended methods of the context-aware of figure incorporation model, comprising the following steps:
(1) the entire music played data of user and the metadata of all music are collected, entire music played data includes
User's history records every broadcasting of music, and the metadata of music includes singer informations, album information and label information.
(2) Heterogeneous Information model is constructed according to the metadata of the entire music played data of all users and all music,
As shown in figure 3, Heterogeneous Information model includes user-music interaction figure, music-music transfer figure and music-metadata knowledge graph.
In user-music interaction figure, user uiWith music mjBetween side right weight beWherein t is user
uiPlay music mjNumber;In music-music transfer figure, music miAnd mjBetween the weight on side beIts
Middle t is music miAnd mjThe number played together;In music-metadata knowledge graph, music miWith metadata djBetween side power
Weight isWherein t is music miIt is endowed metadata djNumber.
(3) Heterogeneous Information figure is handled using figure incorporation model, per song is expressed as feature vector, the figure incorporation model
Objective function Equation are as follows:
Wherein, Em, m indicate all sides in music-music transfer figure,Indicate Em, music m in miAnd mjBetween
Side weight, p (mi,mj) indicate music miAnd mjThe probability played together, is defined as:
Wherein, in which:WithRespectively indicate music miAnd mjFeature vector,TIndicate transposition.
Indicate music miAnd mjIndirect relation power in music-user's interaction figure and music-metadata knowledge graph
Weight, is defined as:
Wherein:Indicate miAdjacent node weight in user-music interaction figure to
Amount,Indicate miAdjacent node weight vectors in music-metadata knowledge graph, cos
() is the cosine similarity of two vectors,Indicate the concatenation of two vectors.
(4) feature vector of per song m can be obtained by minimizing objective function OWherein, there is similar broadcast
The music for putting hereafter (music of the above and below m in play sequence) has similar feature vector, has similar listen to
The music of user or similar metadata has similar feature vector.It herein can be specified according to the requirement to efficiency and accuracy
The dimension of vector, to obtain suitable feature vector, (recommendation results using high-dimensional feature vector are more acurrate, and low dimensional
The computational efficiency of feature vector is higher).
(5) feature vector of the music corresponding to item each in the entire music played data of user record is averaging, and is obtained
To the global interest in music vector of user;From user's entire music played data extract current time for the previous period in
It plays record and forms recent music sequence, and then item each in recent music sequence is played and records corresponding music
Feature vector is averaging, and obtains the context interest in music vector of user.
(6) according to the global interest vector of user and context interest vector, target user u is calculated to music miIt is emerging
Interest, calculation formula are as follows:
Wherein: u indicates target user, miIt is certain song in music libraries;It is the global interest vector of user,
It is the context interest vector of user;It is music miFeature vector, M is music libraries.
(7) all music are ranked up using the calculated result of upper step, top n is recommended target user u, sequence
Calculation formula is as follows:
Wherein: u indicates target user;miAnd mjIt is the music in music libraries;It is the global interest vector of user,
It is the context interest vector of user.
The framework of the music recommended method of context-aware of the present embodiment based on figure incorporation model shown in Fig. 1.It should
Recommended method is divided into two main modulars: preprocessing module and prediction module.In preprocessing module, acquisition user's is all first
Played data and metadata simultaneously construct Heterogeneous Information figure;Figure incorporation model is recycled to extract the feature of music from Heterogeneous Information figure
Vector.In prediction module, record is completely played from target user first and plays the global sound for obtaining user in record in the recent period
Happy interest and context interest in music;Then it is suitable to be recommended according to the global interest of user and context interest to user
Music.
Fig. 2 illustrates the detailed step of user preference prediction, and the complete broadcasting of acquisition user first is recorded and broadcast in the recent period
Put record, and therefrom extract user global interest in music and context interest in music, then using user global interest and
Context interest calculates target user u to music miInterest.
The above-mentioned description to embodiment is for that can understand and apply the invention convenient for those skilled in the art.
Person skilled in the art obviously easily can make various modifications to above-described embodiment, and described herein general
Principle is applied in other embodiments without having to go through creative labor.Therefore, the present invention is not limited to the above embodiments, ability
Field technique personnel announcement according to the present invention, the improvement made for the present invention and modification all should be in protection scope of the present invention
Within.
Claims (1)
1. a kind of context-aware music recommended method based on figure incorporation model, includes the following steps:
(1) entire music for collecting user listens to the label information of sequence and all music;The entire music listens to sequence
Every comprising user's history for music is listened to record, and the label information includes singer, affiliated album and the song of music
Wind type;
(2) metadata of sequence and all music is listened to according to the entire music of all users, is established and is handed over comprising user-music
Mutually figure, music-music transfer figure and music-label knowledge graph Heterogeneous Information model;
It is connected between user and music by sideline in the user-music interaction figure, wherein user u and music mjEdge-stitching
WeighttU, mjMusic m is listened to for user ujNumber, u ∈ U, U indicate the use being made of all users
Family cluster;
It is connected between music and music by sideline in the music-music transfer figure, wherein music miWith music mjEdge-stitching
WeighttMi, mjFor music miWith music mjThe number listened to together, i.e., in same entire music
If listening to music m in sequenceiWith music mjCorresponding listen to is recorded within the scope of certain intervals, then determines music miWith music mj
It is listened to simultaneously;
It is connected between music and label by sideline in the music-label knowledge graph, wherein music mjWith the edge-stitching of label d
WeighttD, mjFor music mjIt is endowed the number of label d, d ∈ D, D expression is made of all labels
Tag library;
(3) according to the following objective function L of the Heterogeneous Information model foundation:
Wherein:Indicate music m in music-music transfer figureiWith music mjEdge-stitching weight,Indicate music mi
With music mjIndirect relation weight, p (mi, mj) indicate music miWith music mjThe probability listened to together, M are indicated by all sounds
The music storehouse of happy composition;
The indirect relation weightExpression formula it is as follows:
Wherein:For music m in user-music interaction figureiAll sideline weights composition vector,For user-music
Music m in interaction figurejAll sideline weights composition vector,For music m in music-label knowledge graphiAll sides
The vector of line weight composition,For music m in music-label knowledge graphjAll sideline weights composition vector, cos (A,
B the cosine similarity of vector A and vector B) are indicated,Indicate the concatenation of two vectors;
Probability p (the mi, mj) expression formula it is as follows:
Wherein:WithRespectively music miWith music mjFeature vector, T indicate vector transposition;
(4) minimum solution is carried out to above-mentioned objective function L, in the hope of the feature vector of per song in music storehouse M;
(5) all listen in sequence is listened to user's entire music and records corresponding musical features vector averaging, obtain user
Global interest in music vector;
(6) recent sound is formed from the record of listening to interior for the previous period that user's entire music listens to extraction current time in sequence
Pleasure listens to sequence;And then all listen in sequence is listened to recent music and records corresponding musical features vector averaging, it obtains
The context interest in music vector of user;
(7) according to the feature vector of per song and the global interest in music vector sum context interest in music vector of user,
User is calculated by the following formula out for the interest value of per song;And then according to interest value to all music in music storehouse from
Small sequence is arrived greatly, and is extracted the maximum several songs of interest value and recommended user;
Wherein: xU, iIt is user u for music miInterest value,WithRespectively music miWith music mjFeature vector, vu
For the global interest in music vector of user u, zuFor the context interest in music vector of user u, T indicates vector transposition, u ∈ U, U
Indicate the user's cluster being made of all users.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610902186.9A CN106528653B (en) | 2016-10-17 | 2016-10-17 | A kind of context-aware music recommended method based on figure incorporation model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610902186.9A CN106528653B (en) | 2016-10-17 | 2016-10-17 | A kind of context-aware music recommended method based on figure incorporation model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106528653A CN106528653A (en) | 2017-03-22 |
CN106528653B true CN106528653B (en) | 2019-04-23 |
Family
ID=58332616
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610902186.9A Active CN106528653B (en) | 2016-10-17 | 2016-10-17 | A kind of context-aware music recommended method based on figure incorporation model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106528653B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108874998B (en) * | 2018-06-14 | 2021-10-19 | 华东师范大学 | Conversational music recommendation method based on mixed feature vector representation |
CN110659382B (en) * | 2019-09-04 | 2021-10-15 | 杭州电子科技大学 | Mixed music recommendation method based on heterogeneous information network representation learning technology |
CN112329928B (en) * | 2020-12-30 | 2021-04-30 | 四川新网银行股份有限公司 | Heterogeneous model-based user satisfaction analysis method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103970873A (en) * | 2014-05-14 | 2014-08-06 | 中国联合网络通信集团有限公司 | Music recommending method and system |
CN105335595A (en) * | 2014-06-30 | 2016-02-17 | 杜比实验室特许公司 | Feeling-based multimedia processing |
CN105677850A (en) * | 2016-01-07 | 2016-06-15 | 浙江大学 | Context-sensing music recommendation method based on neural network model |
CN105975496A (en) * | 2016-04-26 | 2016-09-28 | 清华大学 | Music recommendation method and device based on context sensing |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080027930A1 (en) * | 2006-07-31 | 2008-01-31 | Bohannon Philip L | Methods and apparatus for contextual schema mapping of source documents to target documents |
-
2016
- 2016-10-17 CN CN201610902186.9A patent/CN106528653B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103970873A (en) * | 2014-05-14 | 2014-08-06 | 中国联合网络通信集团有限公司 | Music recommending method and system |
CN105335595A (en) * | 2014-06-30 | 2016-02-17 | 杜比实验室特许公司 | Feeling-based multimedia processing |
CN105677850A (en) * | 2016-01-07 | 2016-06-15 | 浙江大学 | Context-sensing music recommendation method based on neural network model |
CN105975496A (en) * | 2016-04-26 | 2016-09-28 | 清华大学 | Music recommendation method and device based on context sensing |
Non-Patent Citations (1)
Title |
---|
Trust-Based Personalized Service Recommendation: A Network Perspective;Shui-Guang Deng等;《JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY》;20140131;69-80 |
Also Published As
Publication number | Publication date |
---|---|
CN106528653A (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108509534B (en) | Personalized music recommendation system based on deep learning and implementation method thereof | |
CN102024058B (en) | Music recommendation method and system | |
CN105578222B (en) | A kind of information-pushing method and device | |
CN104123315B (en) | The recommendation method and recommendation server of multimedia file | |
CN111083393B (en) | Method for intelligently making short video | |
CN104219575A (en) | Related video recommending method and system | |
WO2015101155A1 (en) | Method for recommending information to user | |
CN106294787A (en) | Information pushing method and device and electronic equipment | |
CN106528653B (en) | A kind of context-aware music recommended method based on figure incorporation model | |
CN104751354B (en) | A kind of advertisement crowd screening technique | |
CN101504834A (en) | Humming type rhythm identification method based on hidden Markov model | |
CN105975496A (en) | Music recommendation method and device based on context sensing | |
CN109376265A (en) | Song recommendations list generation method, medium, device and calculating equipment | |
CN106354860A (en) | Method for automatically labelling and pushing information resource based on label sets | |
CN103226569A (en) | Video providing method, device and system | |
KR101804967B1 (en) | Method and system to recommend music contents by database composed of user's context, recommended music and use pattern | |
CN105608105B (en) | It is a kind of that method is recommended based on the music for listening to context | |
CN105787069A (en) | Personalized music recommendation method | |
US20090271413A1 (en) | Trial listening content distribution system and terminal apparatus | |
Wu | Music personalized recommendation system based on hybrid filtration | |
CN101105943A (en) | Language aided expression system and its method | |
Knees et al. | Music retrieval and recommendation: A tutorial overview | |
CN107818183A (en) | A kind of Party building video pushing method based on three stage combination recommended technologies | |
CN105426550A (en) | Collaborative filtering tag recommendation method and system based on user quality model | |
CN105677850B (en) | A kind of context-aware music recommended method based on neural network model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |