CN108388652A - A kind of method and apparatus for sending song and singly identifying - Google Patents
A kind of method and apparatus for sending song and singly identifying Download PDFInfo
- Publication number
- CN108388652A CN108388652A CN201810169174.9A CN201810169174A CN108388652A CN 108388652 A CN108388652 A CN 108388652A CN 201810169174 A CN201810169174 A CN 201810169174A CN 108388652 A CN108388652 A CN 108388652A
- Authority
- CN
- China
- Prior art keywords
- song
- account
- information
- list
- concern
- 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.)
- Granted
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
- G06F16/63—Querying
- G06F16/638—Presentation of query results
- G06F16/639—Presentation of query results using playlists
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a kind of method and apparatus for sending song and singly identifying, and belong to field of computer technology.The method includes:When detecting that the song list of corresponding first account recommends trigger event to occur, the corresponding account information of the first account is determined;It singly concentrates each song single candidate's song, determines the corresponding song list information of the song list;By the corresponding account information of first account and the single information of the single corresponding song of the song, it is input to the Rating Model trained in advance, predicts first account scoring single to the song;In each song list that candidate song is singly concentrated, determine that the target song of the corresponding default recommendation condition of scoring satisfaction is single;The single song list mark of the target song is sent to the registration terminal of first account.Using the present invention, the single efficiency of selection song can be improved.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of method and apparatus for sending song and singly identifying.
Background technology
With the development of computer technology, the terminals such as mobile phone, computer are widely used, in corresponding terminal
The type of application program is more and more, function is more and more abundant.Playback of songs class application program is a kind of application being in daily use
Program.
User can play song by playback of songs class application program.Currently, playback of songs class application program is often
It is single to provide some songs, wherein include number of songs in song list.User wants to play song by playback of songs class application program
When, it can be in the single-row table of song that playback of songs class application program provides, according to the information such as song single label, title, brief introduction, choosing
The song list for oneself liking listening is selected, it is then possible to click broadcast button, triggering terminal plays the song in selected song list.
In the implementation of the present invention, the inventor finds that the existing technology has at least the following problems:
Based on above-mentioned processing mode, when user wants selection song list every time, it is required to carry in playback of songs class application program
The song list for oneself liking listening is selected in the single-row table of song of confession, the song odd number amount often sung in single-row table is relatively more, to cause to select
Select the less efficient of song list.
Invention content
In order to solve the problems, such as the less efficient of selection song list present in the relevant technologies, an embodiment of the present invention provides one
Kind sends the method and apparatus that song singly identifies.The technical solution is as follows:
In a first aspect, a kind of method for sending song and singly identifying is provided, the method includes:
When detecting that the song list of corresponding first account recommends trigger event to occur, the corresponding account letter of the first account is determined
Breath;
It singly concentrates each song single candidate's song, determines the corresponding song list information of the song list;By first account
The single information of the single corresponding song of corresponding account information and the song, is input to the Rating Model trained in advance, predicts described the
The scoring single to the song of one account;
In each song list that candidate song is singly concentrated, determine that the target song of the corresponding default recommendation condition of scoring satisfaction is single;
The single song list mark of the target song is sent to the registration terminal of first account.
Optionally, described when detecting that the song list of corresponding first account recommends trigger event to occur, determine the first account
Corresponding account information, including:
When detecting that the song list of corresponding first account recommends trigger event to occur, according to the corresponding base of first account
Plinth attribute information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, determine
For the corresponding account information of first account.
Optionally, the corresponding account features information of the determination first account, including:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, determine every in the first concern list of songs
The corresponding song features information of a song;
According to the corresponding song features information of each song in the first concern list of songs, first account is determined
The corresponding account features information in family.
Optionally, the corresponding song features information of each song paid close attention to according to described first in list of songs, really
Determine the corresponding account features information of first account, including:
Determine concern type of first account to each song in the first concern list of songs;And determine institute
State the concern duration of paying close attention to moment apart from current time of first account to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described
The corresponding concern weights of concern type of each song in first concern list of songs;And according to time decay formula and
When first account pays close attention to described first the concern moment of each song in list of songs apart from the concern at current time
It is long, determine the corresponding time dough softening of each song;
By the corresponding concern weights of the concern type of each song time dough softening corresponding with each song
Product calculate each of described first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of the corresponding song features information of song obtains the corresponding account features information of first account.
Optionally, described singly to concentrate each song single candidate's song, it determines the single corresponding song list information of the song, wraps
It includes:
Singly concentrate each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, really
The single characteristic information of the fixed single corresponding song of the song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song
Single single information of corresponding song.
Optionally, the method further includes:
Determine the identification information of each song in library;
The song features gone out that the identification information of each song in the library is input to training in advance extract mould
In type, the song features information of each song in library is determined;
It is single for pre-stored each song, determine the corresponding song features information of each song that the song includes singly;Root
The corresponding song features information of each song for including singly is sung according to described, determines the single characteristic information of the single corresponding song of the song.
Optionally, the method further includes:
For it is pre-stored it is each sing each song for including singly, obtain the identification information of the song and described
The mark of the identification information and the subsequent second preset number song of the song of the first preset number song before song
Information is known, respectively using the identification information of the song as mode input data, by the mark of the first preset number song
The identification information of information and the second preset number song is known as model output data, storage to each mould in training set
In the correspondence of type input data and model output data;
According to the correspondence of each mode input data and model output data in training set, based on by comprising undetermined
The model output data that the song features extraction model of parameter obtains levels off to the model output data in the correspondence
Training philosophy is trained the song features extraction model comprising undetermined parameter, obtains song features extraction model.
Optionally, the method further includes:
According to each account in the training set scoring and each account corresponding account information single to each song, respectively sing single pair
The song list information answered levels off to training set based on the account predicted by Rating Model comprising undetermined parameter single scoring of singing in antiphonal style
In account sing in antiphonal style the training philosophy of single scoring, the Rating Model comprising undetermined parameter is trained, Rating Model is obtained.
Optionally, the method further includes:
For each account, the corresponding concern list of songs of the account is obtained;According to the corresponding concern song of the account
Qu Liebiao counts the number of the song of the different genre types in the corresponding concern list of songs of the account, and described in calculating
The number of the song of different genre types in the corresponding concern list of songs of account is in the corresponding concern song row of the account
Accounting in the total number of song in table obtains the song accounting of the corresponding different genre types of the account;
In each account, the song accounting for choosing corresponding default genre type reaches the target account of default accounting threshold value
Family;
Determine the corresponding account information of target account;
For each song in the corresponding concern list of songs of the target account, the song letter of the song is determined
Breath, by the song information of the account information of the target account and the song, is input to the Rating Model trained in advance, in advance
Survey scoring of the target account to the song;
In each song in the corresponding concern list of songs of the target account, the corresponding maximum third of scoring is determined
Preset number song;
Create the song list being made of the third preset number song.
Second aspect, provides a kind of device for sending song and singly identifying, and described device includes:
First determining module, for when detecting that the song list of corresponding first account recommends trigger event to occur, determining the
The corresponding account information of one account;
Second determining module determines the single corresponding song list letter of the song for singly concentrating each song single candidate's song
Breath;
First prediction module, for singly believing the corresponding account information of first account and the single corresponding song of the song
Breath is input to the Rating Model trained in advance, predicts first account scoring single to the song;
Third determining module, in each song list that candidate song is singly concentrated, determining that corresponding scoring meets default recommend
The target song of condition is single;
Sending module, for sending the single song list mark of the target song to the registration terminal of first account.
Optionally, first determining module, is used for:
When detecting that the song list of corresponding first account recommends trigger event to occur, according to the corresponding base of first account
Plinth attribute information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, determine
For the corresponding account information of first account.
Optionally, first determining module, is used for:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, determine every in the first concern list of songs
The corresponding song features information of a song;
According to the corresponding song features information of each song in the first concern list of songs, first account is determined
The corresponding account features information in family.
Optionally, first determining module, is used for:
Determine concern type of first account to each song in the first concern list of songs;And determine institute
State the concern duration of paying close attention to moment apart from current time of first account to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described
The corresponding concern weights of concern type of each song in first concern list of songs;And according to time decay formula and
When first account pays close attention to described first the concern moment of each song in list of songs apart from the concern at current time
It is long, determine the corresponding time dough softening of each song;
By the corresponding concern weights of the concern type of each song time dough softening corresponding with each song
Product calculate each of described first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of the corresponding song features information of song obtains the corresponding account features information of first account.
Optionally, second determining module, is used for:
Singly concentrate each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, really
The single characteristic information of the fixed single corresponding song of the song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song
Single single information of corresponding song.
Optionally, described device further includes:
4th determining module, the identification information for determining each song in library;
5th determining module, for the identification information of each song in the library to be input to going out for training in advance
Song features extraction model in, determine the song features information of each song in library;
6th determining module, for for pre-stored each song list, determining that each song for including singly of singing corresponds to
Song features information;The corresponding song features information of each song for including singly is sung according to described, determines that the song is single corresponding
The single characteristic information of song.
Optionally, described device further includes:
Memory module, each song for including singly for pre-stored each song, obtains the mark of the song
The identification information of the first preset number song before information and the song and subsequent second present count of the song
The identification information of mesh song, respectively using the identification information of the song as mode input data, by first present count
The identification information of mesh song and the identification information of the second preset number song are as model output data, storage to instruction
In the correspondence for practicing each mode input data and model output data concentrated;
First training module, for the corresponding pass according to each mode input data and model output data in training set
System levels off to the correspondence based on the model output data obtained by the song features extraction model comprising undetermined parameter
In model output data training philosophy, the song features extraction model comprising undetermined parameter is trained, song is obtained
Feature Selection Model.
Optionally, described device further includes:
Second training module, for corresponding to each single scoring of song and each account according to each account in training set
Account information, the single information of the single corresponding song of each song, are sung in antiphonal style single based on the account predicted by the Rating Model comprising undetermined parameter
Scoring level off to the account in training set and sing in antiphonal style the training philosophy of single scoring, the Rating Model comprising undetermined parameter is carried out
Training, obtains Rating Model.
Optionally, described device further includes:
Computing module, for for each account, obtaining the corresponding concern list of songs of the account;According to the account
Corresponding concern list of songs counts the number of the song of the different genre types in the corresponding concern list of songs of the account
Mesh, and the number for calculating the corresponding song for paying close attention to the different genre types in list of songs of the account is corresponded in the account
Concern list of songs in song total number in accounting, the song for obtaining the corresponding different genre types of the account accounts for
Than;
Module is chosen, in each account, the song accounting for choosing corresponding default genre type to reach default accounting
The target account of threshold value;
7th determining module, for determining the corresponding account information of target account;
Second prediction module, for for each song in the corresponding concern list of songs of the target account, determining
The song information of the account information of the target account and the song is input to advance instruction by the song information of the song
The Rating Model practised predicts scoring of the target account to the song;
8th determining module is used in each song in the corresponding concern list of songs of the target account, determining pair
The maximum third preset number song of scoring answered;
Creation module, for creating the song list being made of the third preset number song.
The third aspect provides a kind of server, and the server includes processor and memory, is deposited in the memory
Contain at least one instruction, at least one section of program, code set or instruction set, at least one instruction, at least one section of journey
Sequence, the code set or instruction set are loaded by the processor and are executed to realize the single mark of transmission song as described in relation to the first aspect
Method.
Fourth aspect provides a kind of computer readable storage medium, at least one finger is stored in the storage medium
Enable, at least one section of program, code set or instruction set, at least one instruction, at least one section of program, the code set or
Instruction set is loaded by processor and is executed the method singly identified with the transmission song realized as described in relation to the first aspect.
The advantageous effect that technical solution provided in an embodiment of the present invention is brought is:
In the embodiment of the present invention, when detecting that the song list of corresponding first account recommends trigger event to occur, first is determined
The corresponding account information of account;It singly concentrates each song single candidate's song, determines the song singly corresponding song list information;By first
The single single information of corresponding song of the corresponding account information of account and the song, is input to the Rating Model trained in advance, predicts first
Scoring of the account to the song list;In each song list that candidate song is singly concentrated, determines that corresponding scoring meets and preset recommendation condition
Target song is single;The single song list mark of target song is sent to the registration terminal of the first account.In this way, user can send in server
The single song list mark of target song in, the song list for oneself liking listening is selected, without being selected in all song lists, it is thus possible to carry
The single efficiency of high selection song.
Description of the drawings
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, other are can also be obtained according to these attached drawings
Attached drawing.
Fig. 1 is a kind of method flow diagram for sending song and singly identifying provided in an embodiment of the present invention;
Fig. 2 is a kind of schematic diagram of determining single feature vector of song provided in an embodiment of the present invention;
Fig. 3 is a kind of schematic diagram of determining training sample provided in an embodiment of the present invention;
Fig. 4 is a kind of system framework schematic diagram provided in an embodiment of the present invention;
Fig. 5 is a kind of apparatus structure schematic diagram for sending song and singly identifying provided in an embodiment of the present invention;
Fig. 6 is a kind of apparatus structure schematic diagram for sending song and singly identifying provided in an embodiment of the present invention;
Fig. 7 is a kind of apparatus structure schematic diagram for sending song and singly identifying provided in an embodiment of the present invention;
Fig. 8 is a kind of apparatus structure schematic diagram for sending song and singly identifying provided in an embodiment of the present invention;
Fig. 9 is a kind of apparatus structure schematic diagram for sending song and singly identifying provided in an embodiment of the present invention;
Figure 10 is a kind of structural schematic diagram of server provided in an embodiment of the present invention.
Specific implementation mode
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
An embodiment of the present invention provides a kind of method for sending song and singly identifying, the executive agent of this method is server.Its
In, which can be with the server for sending the single identification function of song, can be the backstage of playback of songs class application program
Server.Server may include the components such as processor, memory, transceiver.Processor can be CPU (Central
Processing Unit, central processing unit) etc., it is determined for the single relevant treatment of target song.Memory can be
RAM (Random Access Memory, random access memory), Flash (flash memory) etc. can be used for storing the number received
According to the data needed for, processing procedure, the data etc. that generate in processing procedure, such as Rating Model.Transceiver can be used for and end
End or other servers carry out data transmission, for example, sending the single song list mark of target song to terminal, transceiver may include day
Line, match circuit, modem etc..
Below in conjunction with specific implementation mode, process flow shown in FIG. 1 is described in detail, content can be as
Under:
Step 101, when detecting that the song list of corresponding first account recommends trigger event to occur, determine that the first account corresponds to
Account information.
Wherein, the corresponding account information of the first account can be the information of the feature or attribute for reflecting the first account.
In force, the song list that corresponding each account can be previously provided in server recommends trigger event, wherein each
The corresponding song of account singly recommends trigger event can be identical, for example the single recommendation trigger event of song can be that preset song singly recommends week
Phase, the corresponding song of each account singly recommend trigger event can also be different, for example, the corresponding song of each account is singly recommended to trigger thing
Part can be the song list recommendation request that the terminal that account is logged in is sent respectively.During the work time, server can be to correspondence
The song list of each account recommends trigger event to be detected, when server detects that the song list of corresponding first account recommends trigger event
When generation (for example, when song list recommendation request that the terminal for receiving the first Account Logon is sent), server may determine that clothes
Current in business device whether to be stored with the target song list to be recommended to the first account, if be stored with, server can be obtained directly
It takes the song list single to the target song of the first account to be recommended to identify, in turn, the registration terminal of the first account can be sent it to,
If be not stored with, it is single that server can determine that the target to be recommended to the first account is sung.Specifically, server can determine
The corresponding account information of first account, wherein the first account can be any account.
Optionally, the corresponding account information of the first account may include account features information and account attributes information, accordingly
, the processing procedure of step 101 can be as follows:When detecting that the song list of corresponding first account recommends trigger event to occur, root
According to the corresponding primary attribute information of the first account and concern song preference data, the corresponding account attributes letter of the first account is determined
Breath;Determine the corresponding account features information of the first account;The corresponding account attributes information of first account and the first account are corresponded to
Account features information, be determined as the corresponding account information of the first account.
Wherein, primary attribute information can be demographic attributes information, may include age information, gender information, close
Note song preference data can be the data for reflecting corresponding account to the degree of concern of all kinds of songs.
In force, when detecting that the song list of corresponding first account recommends trigger event to occur, server can determine
The corresponding account features information of first account, in addition to this, server can also determine the corresponding account attributes letter of the first account
Breath.Specifically, server can obtain the corresponding primary attribute information of the first account, wherein primary attribute information includes the age
Information, gender information etc., and the corresponding concern song preference data of the first account can be obtained (for example, can be the first account
The number of the song of different schools in corresponding first concern list of songs), it in turn, can be by the corresponding basis of the first account
Attribute information and concern song preference data, are determined as the corresponding account attributes information of the first account, can be indicated with m3.It determines
After going out the corresponding account attributes information of the first account and account features information, server can be by the corresponding account category of the first account
Property information and account features information, are determined as the corresponding account information of the first account.In addition, account attributes information can use vector
It indicates, you can be known as account attributes vector, in such cases, when the song list recommendation trigger event hair for detecting corresponding first account
When raw, server can determine the corresponding account attributes vector of the first account, wherein every one-dimensional element of account attributes vector can
To indicate different attributes.Specifically, server can obtain the primary attribute information and concern song preference number of the first account
According in turn, being determined every in account attributes vector according to the primary attribute information of the first account and concern song preference data
The numerical value of element is tieed up to get vectorial to the corresponding account attributes of the first account.
For example, each dimension element of account attributes vector indicate respectively different age bracket (such as be respectively [5,12], [13,
17], [18,24], [25,30], [31,35], [36,40], [41,50], [51,60], be less than 5 and be more than 60, be unknown) and gender
The number of the song of (including man, female, unknown), different schools, in such cases, server gets the first account corresponding year
After age and gender, it may be determined that age and the gender corresponding dimension in account attributes vector in turn can be by the members of the dimension
The numerical value of element is set as 1, other are 0, gets the song of the different schools in the corresponding first concern list of songs of the first account
After bent number, it can will indicate that the corresponding numerical value of the element of the number of different schools is set as the first account corresponding first and closes
The number for noting the song of the different schools in list of songs, to, you can obtain the corresponding account attributes vector of the first account.
Optionally, determine that the processing procedure of the corresponding account features information of the first account can be as follows:Obtain the first account
Corresponding first concern list of songs;In the corresponding song features information of pre-stored each song, determine that the first concern is sung
The corresponding song features information of each song in Qu Liebiao;According to the corresponding song of each song in the first concern list of songs
Bent characteristic information determines the corresponding account features information of the first account.
Wherein, song features information can be the information obtained according to the correlation between song.
In force, the corresponding concern list of songs of each account can be previously stored in server, wherein concern song
List may include account (such as in the past in one month) collection and each song listened in default historical period.When detecting
When the song list of corresponding first account recommends trigger event to occur, server can be sung in the corresponding concern of pre-stored each account
In Qu Liebiao, the corresponding concern list of songs (can be described as the first concern list of songs, can be indicated with U1) of the first account is obtained.
The corresponding song features letter of each song extracted in advance by song features extraction model can also be previously stored in server
Breath, wherein the corresponding song features information of each song can be indicated with vector, you can with referred to as song features vector.It gets
After first concern list of songs, the first concern list of songs can be determined in the corresponding song features information of above-mentioned each song
In the corresponding song features information of each song (can be indicated with i), V can be usediIt indicates, wherein i ∈ U1 in turn can be with
Based on the corresponding song features information of each song determined, the corresponding account features information of the first account, Ke Yiyong are determined
M1 is indicated.That is, the corresponding account features information of the first account can be by each song in the first concern list of songs
Corresponding song features information representation.
Optionally, server can be according to the corresponding each song features information of each song in the first concern list of songs
Weights corresponding with each song features information determine the corresponding account features information of the first account, correspondingly, processing procedure can be with
It is as follows:Determine the concern type of each song in the first account pair first concern list of songs;And determine the first account pair the
Concern duration of the concern moment of each song in one concern list of songs apart from current time;According to pre-stored difference
It pays close attention to type and pays close attention to the correspondence of weights, determine the concern of each song in the first account pair first concern list of songs
The corresponding concern weights of type;And each of list of songs is paid close attention to according to time decay formula and the first account pair first
The concern moment of song apart from the concern duration at current time, determines the corresponding time dough softening of each song;By each song
Corresponding with each song time dough softening of the corresponding concern weights of concern type product, as the corresponding song of each song
The weights of bent characteristic information calculate the weighted sum of the corresponding song features information of each song in the first concern list of songs,
Obtain the corresponding account features information of the first account.
In force, after getting the first concern list of songs, server is in addition to determining in the first concern list of songs
Outside the corresponding song features information of each song, each song in the first account pair first concern list of songs can also be determined
Concern type, wherein concern type can be collection type or listen to type.Different passes can be previously stored in server
It notes type and pays close attention to the correspondence of weights, as shown in table 1.Determine each of the first account pair first concern list of songs
After the concern type of song, each of the first account pair first concern list of songs can be determined in above-mentioned correspondence
The corresponding concern weights of concern type of song (can use α to get to the corresponding concern weights of each songtypeIt indicates).It obtains
After getting the first concern list of songs, server is in addition to determining each song in the first account pair first concern list of songs
It pays close attention to outside the corresponding concern weights of type, can also determine the pass of each song in the first account pair first concern list of songs
The moment is noted, in turn, it may be determined that the concern moment of each song (can use difftime tables apart from the concern duration at current time
Show), obtain the corresponding concern duration of each song.Having time decay formula can be prestored in server, and (wherein, the time declines
It can be e to subtract formula∧- t), the concern moment of each song is determined after the concern duration at current time, it can will be each
The corresponding concern duration of song, band angle of incidence decay formula in (can be using the corresponding concern duration of each song as parameter t
Numerical value), obtaining the corresponding time dough softening of each song (can use βtimeIndicate), wherein concern the moment apart from it is current when
The concern duration at quarter is bigger, and the corresponding time dough softening is smaller.For each song in the first concern list of songs, it is somebody's turn to do
After the corresponding concern weights of the song time dough softening corresponding with the song, it can be multiplied, and product can be determined as
The weights of the corresponding song features information of the song.Obtain the corresponding song features of each song in the first concern list of songs
After the weights of information and the corresponding song features information of each song, each song in the first concern list of songs can be calculated
The weighted sum of corresponding song features information obtains the corresponding account features information of the first account, the i.e. corresponding account of the first account
Family characteristic information m1 can be ΣU1αtypeβtimeVi。
Table 1
Pay close attention to type | Pay close attention to weights |
Collect type | 1 |
Listen to type | 0.5 |
Step 102, it singly concentrates each song single candidate's song, determines the song singly corresponding song list information;By the first account
The single single information of corresponding song of the corresponding account information in family and the song, is input to the Rating Model trained in advance, the first account of prediction
Scoring of the family to the song list.
In force, when detecting that the song list of corresponding first account recommends trigger event to occur, single collection is sung for candidate
Each of song it is single, server can also determine that candidate song singly concentrates the single corresponding song list information of each song.Determine first
After the corresponding account information of account and candidate song singly concentrate the single corresponding song list information of each song, the first account pair can be predicted
Candidate's song singly concentrates the scoring of each song list.Specifically, the scoring mould trained in advance can be previously stored in server
Type, wherein the input of Rating Model can be the song information of the account information and certain song of certain account, and output can be the account
The scoring of the family song, Rating Model can be neural network Rating Models.It singly concentrates each song single candidate's song, services
The single single information of corresponding song of the corresponding account information of first account and the song can be input in Rating Model by device, predict the
Scoring of one account to the song list.
Optionally, the training process of above-mentioned Rating Model can be as follows:Single to each song according to each account in training set
Scoring and the single corresponding song list information of the corresponding account information of each account, each song, based on passing through the scoring comprising undetermined parameter
The account of model prediction sing in antiphonal style single scoring level off to the account in training set sing in antiphonal style list scoring training philosophy, to comprising waiting for
The Rating Model for determining parameter is trained, and obtains Rating Model.
In force, training set and the Rating Model comprising undetermined parameter can be previously stored in server (can claim
For Rating Model to be trained), wherein each training sample in training set can be the scoring single to each song of each account,
And the single corresponding song list information of the account corresponding account information and the song, training sample may include positive sample and negative sample
This, Rating Model to be trained may include that (wherein, input layer is the corresponding account information of account, song single pair to an input layer
The song list information answered), sensing layer 1 (wherein, sensing layer 1 may include 1024 neurons, activation primitive relu), sensing layer
2 (wherein, sensing layer 2 may include 512 neurons, activation primitive relu), (wherein, sensing layer 3 can wrap sensing layer 3
Include 256 neurons, activation primitive relu) and an output layer (activation primitive sigmoid).For every in training set
A training sample, server can be defeated by the single information of the corresponding account information of account song corresponding with song list in the training sample
Enter into the Rating Model comprising undetermined parameter, obtains the account singing in antiphonal style list predicted by the Rating Model comprising undetermined parameter
Scoring, in turn, can be sung in antiphonal style single scoring based on the account predicted by Rating Model comprising undetermined parameter, level off to the instruction
Practice the sample account to the training philosophy of the scoring of the song list, treats trained Rating Model and be trained.All training samples are complete
Portion's training finishes, you can obtains final Rating Model, you can the Rating Model after being trained, for example, packet can will be passed through
The scoring of the account singing in antiphonal style list of Rating Model prediction containing undetermined parameter, the difference for single scoring of singing in antiphonal style with the account in training set
As object function, it is based on gradient descent method, which is trained (for example, the object function can be sought most
Small value), the parameter value of undetermined parameter is obtained to get to final Rating Model.In addition, may not be account in training sample
The single scoring of singing in antiphonal style, when certain account paid close attention to certain song list, the account can be 1 to the scoring of the song list, be otherwise 0.
Optionally, it may include the single characteristic information of the single corresponding song of song and the single attribute information of song to sing single information, correspondingly, step
Rapid 102 processing procedure can be as follows:Singly concentrate each song single candidate's song, in the corresponding song of pre-stored each song list
In single characteristic information, the song singly single characteristic information of corresponding song is determined;It is concerned information according to song list is corresponding, determines song single pair
The song list attribute information answered;By the single single corresponding song list characteristic information of the single attribute information of corresponding song and the song of the song, it is determined as
The single single information of corresponding song of the song.
Wherein, it can be for reflecting that corresponding song is singly concerned the information of situation to be concerned information, and being concerned information can
To include by information such as amount of collection, playback volumes, it can be the information for reflecting the corresponding single attribute of song to sing single attribute information.
In force, the single characteristic information of the single corresponding song of each song can be previously stored in server, wherein respectively sing single pair
The song list characteristic information answered can be indicated with vector, you can be known as singing single feature vector.Each of candidate's song is singly concentrated
Song is single, and server can determine the song singly corresponding song Dan Te in the single characteristic information of the single corresponding song of pre-stored each song
Reference ceases (can be indicated with m2).
It can be previously stored with that each song is single corresponding to be concerned information in server.The each song singly concentrated for candidate song
Single, server can be concerned in information each song is singly corresponding, is determined that the song is singly corresponding and is concerned information, in turn, can be with
According to information is concerned, determine the song singly the single attribute information of corresponding song (for example, letter can be concerned by the song is singly corresponding
Breath, single attribute information should respectively be sung by being determined as the song single pair), it can be indicated with m4, wherein the single attribute information of song can use vector table
Show, alternatively referred to as sings single attribute vector.In addition, the single attribute information of song can also include the single primary attribute information of song, wherein base
Plinth attribute information can be the information for reflecting the corresponding single primary attribute of song, for example, primary attribute information may include song
The number for the song for including singly, the number etc. for singing the song of different schools in list, i.e. server can will sing singly corresponding closed
Information and primary attribute information are noted, the song singly single attribute information of corresponding song is determined as.
It is the case where singing single attribute vector for the single attribute information of song, singing the often dimension element of single attribute vector can indicate pre-
If different attribute (for example, preset different attribute may include amount of collection, the number of listening the song that song amount, song include singly,
The number of song etc. of different schools).In such cases, each song is singly concentrated singly for candidate's song, server is got
The song singly it is corresponding be concerned information and primary attribute information after, can according to the song singly it is corresponding be concerned information and basis belong to
Property information, determine the song singly single attribute vector of corresponding song.Each song is singly concentrated singly for song, determines that each song single pair is answered
Song list attribute information and the song singly after the single characteristic information of corresponding song, can by the single corresponding song list attribute information of each song be somebody's turn to do
The single characteristic information of the single corresponding song of song is determined as the song singly single information of corresponding song.
Optionally, pre-stored song features information can be that server first passes through song features extraction model and obtains in advance
, singing single characteristic information can be that server is obtained by singing the corresponding song features information of song for including singly, correspondingly,
Processing procedure can be as follows:Determine the identification information of each song in library;By the mark of each song in library
In information input to the song features extraction model of training gone out in advance, the song features letter of each song in library is determined
Breath;It is single for pre-stored each song, determine the corresponding song features information of each song that the song includes singly;According to the song list
Including the corresponding song features information of each song, determine the song singly single characteristic information of corresponding song.
In force, each song can be previously stored in library, for each song in library, server can be with
It is encoded, the identification information of each song is obtained, wherein identification information can be mark vector, the mark of each song
The dimension of vector is the total quantity of the song in library, and the numerical value of one-dimensional element is in the corresponding mark vector of each song
1, the numerical value of other dimension elements is 0, and the dimension that each mark vector numerical value is 1 is different.For example, the sum of the song in library
Amount is n1, then the mark vector of each song can be the vector of n1 dimensions, the mark vector of first song can be (1,0,
0 ..., 0,0), the mark vector of second song can be (0,1,0 ..., 0,0) and so on, it obtains every in library
The mark vector of a song.
For each song in library, after the identification information for determining the song, the mark of the song can be believed
Breath is input in the song features extraction model trained in advance, wherein song features extraction model can be neural network mould
Type can be the Computational frame of item2vec, and in turn, server can hide the last one in song features extraction model
The output of layer is as a result, be determined as the corresponding song features information of the song.Server can be sung according to above-mentioned processing mode
The corresponding song features information of each song in Qu Ku, and stored.Determine the corresponding song features of each song
After information, server can determine the single song Dan Te of pre-stored each song according to the corresponding song features information of each song
Reference ceases.Specifically, single for pre-stored each song, server can determine the song that the song includes singly, in turn, can be with
In the corresponding song features information of each song, the corresponding song features information of song that the song includes singly is determined.Determine this
After singing the corresponding song features information of song for including singly, the corresponding song features information of song that the song includes singly can be calculated
Average value, in turn, the average value that can will be obtained is determined as the song singly single characteristic information of corresponding song, wherein song features
It can be the single feature vector of song that information, which can be that song features are vectorial, sing single characteristic information,.For example, song list A includes song 1, song
Song 2, song 3,1 corresponding song features vector of song be (99,170,50), 2 corresponding song features vector of song be (70,
199,50), 3 corresponding song features vector of song is (80,180,50), then, can for the often dimension element of the single feature vector of song
To calculate the average value of the numerical value of the dimension element in the corresponding song features vector of song that song includes singly, obtain singing single feature to
The numerical value of the dimension element of amount, you can to obtain singing single feature vector for (83,183,50), as shown in Figure 2.
Optionally, above-mentioned song features extraction model can be that server trains in advance, correspondingly, processing procedure can
With as follows:For each song that pre-stored each song includes singly, obtain before the identification information and song of song
The identification information of the identification information and the subsequent second preset number song of song of first preset number song, respectively should
The identification information of song is as mode input data, by the identification information of the first preset number song and the second preset number
The identification information of song is stored to each mode input data and model output data in training set as model output data
In correspondence;According to the correspondence of each mode input data and model output data in training set, based on by comprising
The model output data that the song features extraction model of undetermined parameter obtains levels off to the model output data in correspondence
Training philosophy is trained the song features extraction model comprising undetermined parameter, obtains song features extraction model.
In force, before training song features extraction model, server can determine each instruction in training set first
Practice sample.Specifically, obtaining, pre-stored each song is single, and for each song in each song list, server can obtain
The identification information and the song of the first preset number song in the identification information of the song and the song list before the song
In the identification information of the subsequent second preset number song of the song in list, in turn, the mark of the song can be believed respectively
Breath is believed as mode input data, by the mark of the identification information of the first preset number song and the second preset number song
Breath is used as model output data, in storage to the correspondence of mode input data and model output data in training set,
In, the first preset number can be identical with the second preset number, can not also be identical.For example, the first preset number and second is in advance
If number is 2, the song that song list A includes is:Song 1, song 2, song 3, song 4, song 5, song 6, for song 4, clothes
Be engaged in device can using the identification information of song 4 be used as mode input data, song 2, song 3, song 5, song 6 identification information work
It is stored to the mode input data and model output data in training set as a training sample for model output data
In correspondence, as shown in Figure 3.
There is certain correlation, thus, it is possible to pass through the adjacent song in same song list between song in same song list
Bent correlation, training song features extraction model.Specifically, after obtaining training sample, for each training in training set
Mode input data in the training sample can be input to the extraction mould of the song features comprising undetermined parameter by sample, server
In type (can be described as song features extraction model to be trained), obtain determining the model by song features extraction model to be trained
The corresponding model output data of input data (including wherein, undetermined parameter in the model output data) can be based on logical in turn
It is defeated to cross the model that the model output data that the song features extraction model comprising undetermined parameter obtains levels off in correspondence
Enter the training philosophy of the corresponding model output data of data (for example, can will be extracted by the song features comprising undetermined parameter
The model output data that model obtains, the difference of model output data corresponding with the mode input data in correspondence,
As object function, in turn, the minimum value of object function can be sought), it treats trained song features extraction model and is trained.
At the end of waiting for all training sample training, you can obtain the trained values of undetermined parameter, can also obtain final song features
Extraction model, wherein song features extraction model can be neural network model.
Optionally, it is single can also to create song by Rating Model for server, wherein the song of establishment singly may be used to provide song
The training sample of bent Feature Selection Model also can be used as the song list in Candidate Set, correspondingly, processing procedure can be as follows:For
Each account obtains the account corresponding concern list of songs;According to the account corresponding concern list of songs, the account is counted
The number of the song of different genre types in corresponding concern list of songs, and calculate the account corresponding concern list of songs
In different genre types song number the account it is corresponding concern list of songs in song total number in account for
Than obtaining the song accounting of the account corresponding different genre types;In each account, corresponding default genre type is chosen
Song accounting reaches the target account of default accounting threshold value;Determine the corresponding account information of target account;For target account pair
Each song in the concern list of songs answered, determines the song information of the song, by the account information of the target account and is somebody's turn to do
The song information of song is input to the Rating Model trained in advance, scoring of the prediction target account to the song;In target account
In each song in the corresponding concern list of songs in family, the corresponding maximum third preset number song of scoring is determined;It creates
The song list being made of third preset number song.
In force, pre-stored song can be singly that server is created by Rating Model in server, that is,
It says, server can create the song list of certain default genre type by Rating Model.To create the song list for presetting genre type, clothes
Business device can count the song accounting of the corresponding different genre types of each account.Specifically, for each account, server can be with
The account corresponding concern list of songs is obtained, in turn, the different genre types that concern list of songs includes can be counted
The number of song, and calculate different genre types song number the concern list of songs song that includes total number
Accounting obtains the song accounting of the account corresponding different genre types.For example, being wrapped in 1 corresponding concern list of songs 1 of account
It includes:Song 1 (wherein, corresponding genre type be x), song 2 (wherein, corresponding genre type be x), song 3 it is (wherein, right
The genre type answered be x), song 4 (wherein, corresponding genre type be y), song 5 (wherein, corresponding genre type be x),
Then the number of the song of the genre type x in the corresponding concern list of songs of server statistics account 1 is 4,1 corresponding pass of account
The number for noting the song of the genre type y in list of songs is 1, correspondingly, can be obtained 1 corresponding genre type x's of account
Song accounting be 0.8 (4/5), genre type y song accounting be 0.2 (1/5).Obtain the corresponding different genre types of each account
Song accounting after, server can be in each account, and choosing corresponding default genre type, (wherein, default genre type can
Think one, or multiple) song accounting reach the account of default accounting threshold value and (can be described as target account, wherein mesh
It can be one to mark account, can also be multiple).For example, the song accounting of 1 corresponding genre type x of account is 0.8, account 2
The song accounting of corresponding genre type x is 0.6, the song accounting of 3 corresponding genre type x of account is 0.4, presets school class
Type is x genre types, and it is 0.5 to preset accounting threshold value, then account 1 and account 2 can be determined as target account by server.
After determining target account, server can determine target account to the corresponding pass of target account by Rating Model
Note the scoring of each song in list of songs.Specifically, server can be according to the corresponding account information of determining first account
Processing mode, determine the corresponding account information of target account.In addition, in the corresponding concern list of songs of target account
Each song, server can determine the corresponding song of the song according to the processing mode for determining the single corresponding single information of song of song
The song information of the account information of target account and the song can be input in Rating Model by information in turn, predict target
Scoring of the account to the song.Obtain the scoring of each song in target account concern list of songs corresponding to target account
Afterwards, server can choose corresponding scoring maximum the in each song in the corresponding concern list of songs of target account
Three preset numbers (for example, third preset number is 20) a song can create the third preset number by selecting in turn
The song list that song is constituted.
Step 103, in each song list that candidate song is singly concentrated, determine that corresponding scoring meets the target for presetting recommendation condition
Song is single.
Wherein, it can be for judging whether the song list is pushed away according to the single scoring of the first account singing in antiphonal style to preset recommendation condition
The condition recommended.
In force, it after server determines the single scoring of each song that the first account singly concentrates candidate song, can wait
In each song list that choosing song is singly concentrated, determine that corresponding scoring meets the song list (it is single to can be described as target song) for presetting recommendation condition.Tool
Body, server can determine the corresponding highest preset number target song of scoring in each song list that candidate song is singly concentrated
It is single, alternatively, server can determine that corresponding scoring is more than the mesh of default scoring threshold value in each song list that candidate song is singly concentrated
Mark song is single.
Step 104, the single song list mark of target song is sent to the registration terminal of the first account.
In force, after server determines target song list, it is single target song can be sent to the registration terminal of the first account
Song list mark, correspondingly, terminal can receive the single song list mark of target song of server transmission, and it can be shown
Show, so that user selects the song list for oneself liking listening, as shown in Figure 4.
In the embodiment of the present invention, when detecting that the song list of corresponding first account recommends trigger event to occur, first is determined
The corresponding account information of account;It singly concentrates each song single candidate's song, determines the song singly corresponding song list information;By first
The single single information of corresponding song of the corresponding account information of account and the song, is input to the Rating Model trained in advance, predicts first
Scoring of the account to the song list;In each song list that candidate song is singly concentrated, determines that corresponding scoring meets and preset recommendation condition
Target song is single;The single song list mark of target song is sent to the registration terminal of the first account.In this way, user can send in server
The single song list mark of target song in, the song list for oneself liking listening is selected, without being selected in all song lists, it is thus possible to carry
The single efficiency of high selection song.
Based on the same technical idea, the embodiment of the present invention additionally provides a kind of device for sending song and singly identifying, such as Fig. 5 institutes
Show, which includes:
First determining module 510, for when detecting that the song list of corresponding first account recommends trigger event to occur, determining
The corresponding account information of first account;
Second determining module 520 determines the corresponding song list of the song list for singly concentrating each song single candidate's song
Information;
First prediction module 530, for the corresponding account information of first account and the single corresponding song of the song is single
Information is input to the Rating Model trained in advance, predicts first account scoring single to the song;
Third determining module 540, in each song list that candidate song is singly concentrated, determining that corresponding scoring meets default push away
The target song for recommending condition is single;
Sending module 550, for sending the single song list mark of the target song to the registration terminal of first account.
Optionally, first determining module 510, is used for:
When detecting that the song list of corresponding first account recommends trigger event to occur, according to the corresponding base of first account
Plinth attribute information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, determine
For the corresponding account information of first account.
Optionally, first determining module 510, is used for:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, determine every in the first concern list of songs
The corresponding song features information of a song;
According to the corresponding song features information of each song in the first concern list of songs, first account is determined
The corresponding account features information in family.
Optionally, first determining module 510, is used for:
Determine concern type of first account to each song in the first concern list of songs;And determine institute
State the concern duration of paying close attention to moment apart from current time of first account to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described
The corresponding concern weights of concern type of each song in first concern list of songs;And according to time decay formula and
When first account pays close attention to described first the concern moment of each song in list of songs apart from the concern at current time
It is long, determine the corresponding time dough softening of each song;
By the corresponding concern weights of the concern type of each song time dough softening corresponding with each song
Product calculate each of described first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of the corresponding song features information of song obtains the corresponding account features information of first account.
Optionally, second determining module 520, is used for:
Singly concentrate each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, really
The single characteristic information of the fixed single corresponding song of the song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song
Single single information of corresponding song.
Optionally, as shown in fig. 6, described device further includes:
4th determining module 560, the identification information for determining each song in library;It will be in the library
The identification information of each song is input in the song features extraction model of training gone out in advance, determines each song in library
Bent song features information;It is single for pre-stored each song, determine that the corresponding song of each song that the song includes singly is special
Reference ceases;The corresponding song features information of each song for including singly is sung according to described, determines the single feature of the single corresponding song of the song
Information.
Optionally, as shown in fig. 7, described device further includes:
Memory module 570, each song for including singly for pre-stored each song, obtains the mark of the song
Identification information and the song subsequent second for knowing the first preset number song before information and the song are default
The identification information of number song is preset respectively using the identification information of the song as mode input data by described first
As model output data, storage is arrived for the identification information of number song and the identification information of the second preset number song
In the correspondence of each mode input data and model output data in training set;
First training module 580, for the correspondence according to each mode input data and model output data in training set
Relationship levels off to described corresponding close based on the model output data obtained by the song features extraction model comprising undetermined parameter
The training philosophy of model output data in system is trained the song features extraction model comprising undetermined parameter, is sung
Bent Feature Selection Model.
Optionally, as shown in figure 8, described device further includes:
Second training module 590, for being corresponded to each single scoring of song and each account according to each account in training set
Account information, the single information of the single corresponding song of each song, based on the account singing in antiphonal style predicted by the Rating Model comprising undetermined parameter
Single scoring level off to the account in training set sing in antiphonal style list scoring training philosophy, to the Rating Model comprising undetermined parameter into
Row training, obtains Rating Model.
Optionally, as shown in figure 9, described device further includes:
Computing module 5100, for for each account, obtaining the corresponding concern list of songs of the account;According to described
The corresponding concern list of songs of account counts the song of the different genre types in the corresponding concern list of songs of the account
Number, and the number of the corresponding song for paying close attention to the different genre types in list of songs of the account is calculated in the account pair
Accounting in the total number of song in the concern list of songs answered obtains the song of the corresponding different genre types of the account
Accounting;
Module 5110 is chosen, in each account, the song accounting for choosing corresponding default genre type to reach default
The target account of accounting threshold value;
5th determining module 5120, for determining the corresponding account information of target account;
Second prediction module 5130 is used for for each song in the corresponding concern list of songs of the target account,
The song information of the account information of the target account and the song is input to pre- by the song information for determining the song
The Rating Model first trained predicts scoring of the target account to the song;
6th determining module 5140 is used in each song in the corresponding concern list of songs of the target account, really
The fixed corresponding maximum third preset number song of scoring;
Creation module 5150, for creating the song list being made of the third preset number song.
In the embodiment of the present invention, when detecting that the song list of corresponding first account recommends trigger event to occur, first is determined
The corresponding account information of account;It singly concentrates each song single candidate's song, determines the song singly corresponding song list information;By first
The single single information of corresponding song of the corresponding account information of account and the song, is input to the Rating Model trained in advance, predicts first
Scoring of the account to the song list;In each song list that candidate song is singly concentrated, determines that corresponding scoring meets and preset recommendation condition
Target song is single;The single song list mark of target song is sent to the registration terminal of the first account.In this way, user can send in server
The single song list mark of target song in, the song list for oneself liking listening is selected, without being selected in all song lists, it is thus possible to carry
The single efficiency of high selection song.
It should be noted that:The transmission device that singly identifies of song that above-described embodiment provides send sing single mark when, only with
The division progress of above-mentioned each function module, can be as needed and by above-mentioned function distribution by not for example, in practical application
Same function module is completed, i.e., the internal structure of server is divided into different function modules, described above complete to complete
Portion or partial function.In addition, the method that device and transmission song that the transmission song that above-described embodiment provides singly identifies singly identify is real
It applies example and belongs to same design, specific implementation process refers to embodiment of the method, and which is not described herein again.
Figure 10 is the structural schematic diagram of server provided in an embodiment of the present invention.The server 1900 can be because of configuration or performance
It is different and generate bigger difference, may include one or more central processing units (central processing
Units, CPU) 1922 (for example, one or more processors) and memory 1932, one or more storage applications
The storage medium 1930 (such as one or more mass memory units) of program 1942 or data 1944.Wherein, memory
1932 and storage medium 1930 can be of short duration storage or persistent storage.The program for being stored in storage medium 1930 may include one
A or more than one module (diagram does not mark), each module may include to the series of instructions operation in server.More into
One step, central processing unit 1922 could be provided as communicating with storage medium 1930, and storage medium is executed on server 1900
Series of instructions operation in 1930.
Server 1900 can also include one or more power supplys 1926, one or more wired or wireless nets
Network interface 1950, one or more input/output interfaces 1958, one or more keyboards 1956, and/or, one or
More than one operating system 1941, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM
Etc..
Server 1900 may include have memory and one either more than one program one of them or one
A procedure above is stored in memory, and be configured to by one either more than one processor execute it is one or one
A procedure above includes the instruction for being operated below:
When detecting that the song list of corresponding first account recommends trigger event to occur, the corresponding account letter of the first account is determined
Breath;
It singly concentrates each song single candidate's song, determines the corresponding song list information of the song list;By first account
The single information of the single corresponding song of corresponding account information and the song, is input to the Rating Model trained in advance, predicts described the
The scoring single to the song of one account;
In each song list that candidate song is singly concentrated, determine that the target song of the corresponding default recommendation condition of scoring satisfaction is single;
The single song list mark of the target song is sent to the registration terminal of first account.
Optionally, described when detecting that the song list of corresponding first account recommends trigger event to occur, determine the first account
Corresponding account information, including:
When detecting that the song list of corresponding first account recommends trigger event to occur, according to the corresponding base of first account
Plinth attribute information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, determine
For the corresponding account information of first account.
Optionally, the corresponding account features information of the determination first account, including:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, determine every in the first concern list of songs
The corresponding song features information of a song;
According to the corresponding song features information of each song in the first concern list of songs, first account is determined
The corresponding account features information in family.
Optionally, the corresponding song features information of each song paid close attention to according to described first in list of songs, really
Determine the corresponding account features information of first account, including:
Determine concern type of first account to each song in the first concern list of songs;And determine institute
State the concern duration of paying close attention to moment apart from current time of first account to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described
The corresponding concern weights of concern type of each song in first concern list of songs;And according to time decay formula and
When first account pays close attention to described first the concern moment of each song in list of songs apart from the concern at current time
It is long, determine the corresponding time dough softening of each song;
By the corresponding concern weights of the concern type of each song time dough softening corresponding with each song
Product calculate each of described first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of the corresponding song features information of song obtains the corresponding account features information of first account.
Optionally, described singly to concentrate each song single candidate's song, it determines the single corresponding song list information of the song, wraps
It includes:
Singly concentrate each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, really
The single characteristic information of the fixed single corresponding song of the song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song
Single single information of corresponding song.
Optionally, the method further includes:
Determine the identification information of each song in library;
The song features gone out that the identification information of each song in the library is input to training in advance extract mould
In type, the song features information of each song in library is determined;
It is single for pre-stored each song, determine the corresponding song features information of each song that the song includes singly;Root
The corresponding song features information of each song for including singly is sung according to described, determines the single characteristic information of the single corresponding song of the song.
Optionally, the method further includes:
For it is pre-stored it is each sing each song for including singly, obtain the identification information of the song and described
The mark of the identification information and the subsequent second preset number song of the song of the first preset number song before song
Information is known, respectively using the identification information of the song as mode input data, by the mark of the first preset number song
The identification information of information and the second preset number song is known as model output data, storage to each mould in training set
In the correspondence of type input data and model output data;
According to the correspondence of each mode input data and model output data in training set, based on by comprising undetermined
The model output data that the song features extraction model of parameter obtains levels off to the model output data in the correspondence
Training philosophy is trained the song features extraction model comprising undetermined parameter, obtains song features extraction model.
Optionally, the method further includes:
According to each account in the training set scoring and each account corresponding account information single to each song, respectively sing single pair
The song list information answered levels off to training set based on the account predicted by Rating Model comprising undetermined parameter single scoring of singing in antiphonal style
In account sing in antiphonal style the training philosophy of single scoring, the Rating Model comprising undetermined parameter is trained, Rating Model is obtained.
Optionally, the method further includes:
For each account, the corresponding concern list of songs of the account is obtained;According to the corresponding concern song of the account
Qu Liebiao counts the number of the song of the different genre types in the corresponding concern list of songs of the account, and described in calculating
The number of the song of different genre types in the corresponding concern list of songs of account is in the corresponding concern song row of the account
Accounting in the total number of song in table obtains the song accounting of the corresponding different genre types of the account;
In each account, the song accounting for choosing corresponding default genre type reaches the target account of default accounting threshold value
Family;
Determine the corresponding account information of target account;
For each song in the corresponding concern list of songs of the target account, the song letter of the song is determined
Breath, by the song information of the account information of the target account and the song, is input to the Rating Model trained in advance, in advance
Survey scoring of the target account to the song;
In each song in the corresponding concern list of songs of the target account, the corresponding maximum third of scoring is determined
Preset number song;
Create the song list being made of the third preset number song.
In the embodiment of the present invention, when detecting that the song list of corresponding first account recommends trigger event to occur, first is determined
The corresponding account information of account;It singly concentrates each song single candidate's song, determines the song singly corresponding song list information;By first
The single single information of corresponding song of the corresponding account information of account and the song, is input to the Rating Model trained in advance, predicts first
Scoring of the account to the song list;In each song list that candidate song is singly concentrated, determines that corresponding scoring meets and preset recommendation condition
Target song is single;The single song list mark of target song is sent to the registration terminal of the first account.In this way, user can send in server
The single song list mark of target song in, the song list for oneself liking listening is selected, without being selected in all song lists, it is thus possible to carry
The single efficiency of high selection song.
One of ordinary skill in the art will appreciate that realizing that all or part of step of above-described embodiment can pass through hardware
It completes, relevant hardware can also be instructed to complete by program, the program can be stored in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only memory, disk or CD etc..
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention.
Claims (20)
1. a kind of method for sending song and singly identifying, which is characterized in that the method includes:
When detecting that the song list of corresponding first account recommends trigger event to occur, the corresponding account information of the first account is determined;
It singly concentrates each song single candidate's song, determines the corresponding song list information of the song list;
By the single corresponding song list information of the corresponding account information of first account and the song, it is input to commenting of training in advance
Sub-model predicts first account scoring single to the song;
In each song list that candidate song is singly concentrated, determine that the target song of the corresponding default recommendation condition of scoring satisfaction is single;
The single song list mark of the target song is sent to the registration terminal of first account.
2. according to the method described in claim 1, it is characterized in that, described ought detect that the song list of corresponding first account is recommended to touch
When hair event occurs, the corresponding account information of the first account is determined, including:
When detecting that the song list of corresponding first account recommends trigger event to occur, belonged to according to the corresponding basis of first account
Property information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, it is determined as institute
State the corresponding account information of the first account.
3. according to the method described in claim 2, it is characterized in that, the corresponding account features letter of the determination first account
Breath, including:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, each song in the first concern list of songs is determined
Bent corresponding song features information;
According to the corresponding song features information of each song in the first concern list of songs, first account pair is determined
The account features information answered.
4. according to the method described in claim 3, it is characterized in that, described pay close attention to each of list of songs according to described first
The corresponding song features information of song determines the corresponding account features information of first account, including:
Determine concern type of first account to each song in the first concern list of songs;And determine described the
One account pays close attention to concern duration of the moment apart from current time to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described first
Pay close attention to the corresponding concern weights of concern type of each song in list of songs;And according to time decay formula and described
First account pays close attention to concern duration of the moment apart from current time to each song in the first concern list of songs, really
Determine the corresponding time dough softening of each song;
By multiplying for corresponding concern weights and each corresponding time dough softening of song of the concern type of each song
Product calculates each song in the first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of corresponding song features information obtains the corresponding account features information of first account.
5. determining according to the method described in claim 1, it is characterized in that, described singly concentrate each song single candidate's song
The single information of the single corresponding song of the song, including:
It singly concentrates each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, determines institute
State the single characteristic information of the single corresponding song of song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song single pair
The song list information answered.
6. the method according to claim 3 or 5, which is characterized in that the method further includes:
Determine the identification information of each song in library;
The identification information of each song in the library is input in the song features extraction model of training gone out in advance,
Determine the song features information of each song in library;
It is single for pre-stored each song, determine the corresponding song features information of each song that the song includes singly;According to institute
The corresponding song features information of each song that song includes singly is stated, determines the single characteristic information of the single corresponding song of the song.
7. according to the method described in claim 6, it is characterized in that, the method further includes:
For each song that pre-stored each song includes singly, the identification information of the song and the song are obtained
The identification information of first preset number song of front is believed with the mark of the subsequent second preset number song of the song
Breath believes the mark of the first preset number song respectively using the identification information of the song as mode input data
As model output data, storage is defeated to each model in training set for breath and the identification information of the second preset number song
In the correspondence for entering data and model output data;
According to the correspondence of each mode input data and model output data in training set, based on by including undetermined parameter
The obtained model output data of song features extraction model level off to the training of the model output data in the correspondence
Principle is trained the song features extraction model comprising undetermined parameter, obtains song features extraction model.
8. according to the method described in claim 1, it is characterized in that, the method further includes:
It is single corresponding according to each account in the training set scoring and each account corresponding account information single to each song, each song
The single information of song, the scoring based on the account singing in antiphonal style list predicted by the Rating Model comprising undetermined parameter level off in training set
The training philosophy of the single scoring of account singing in antiphonal style, is trained the Rating Model comprising undetermined parameter, obtains Rating Model.
9. method according to claim 1 or claim 7, which is characterized in that the method further includes:
For each account, the corresponding concern list of songs of the account is obtained;According to the corresponding concern song row of the account
Table, counts the number of the song of the different genre types in the corresponding concern list of songs of the account, and calculates the account
The number of the song of different genre types in corresponding concern list of songs is in the corresponding concern list of songs of the account
Song total number in accounting, obtain the song accounting of the corresponding different genre types of the account;
In each account, the song accounting for choosing corresponding default genre type reaches the target account of default accounting threshold value;
Determine the corresponding account information of target account;
For each song in the corresponding concern list of songs of the target account, the song information of the song is determined, it will
The song information of the account information of the target account and the song is input to the Rating Model trained in advance, predicts institute
State scoring of the target account to the song;
In each song in the corresponding concern list of songs of the target account, determine that the corresponding maximum third of scoring is default
Number song;
Create the song list being made of the third preset number song.
10. a kind of sending the device that singly identifies of song, which is characterized in that described device includes:
First determining module, for when detecting that the song list of corresponding first account recommends trigger event to occur, determining the first account
The corresponding account information in family;
Second determining module determines the corresponding song list information of the song list for singly concentrating each song single candidate's song;
First prediction module, for will the single corresponding song list information of the corresponding account information of first account and the song, it is defeated
Enter to the Rating Model trained in advance, predicts first account scoring single to the song;
Third determining module presets recommendation condition in each song list that candidate song is singly concentrated, determining that corresponding scoring meets
Target song it is single;
Sending module, for sending the single song list mark of the target song to the registration terminal of first account.
11. device according to claim 10, which is characterized in that first determining module is used for:
When detecting that the song list of corresponding first account recommends trigger event to occur, belonged to according to the corresponding basis of first account
Property information and concern song preference data, determine the corresponding account attributes information of first account;
Determine the corresponding account features information of first account;
By the corresponding account attributes information of first account and the corresponding account features information of first account, it is determined as institute
State the corresponding account information of the first account.
12. according to the devices described in claim 11, which is characterized in that first determining module is used for:
Obtain the corresponding first concern list of songs of first account;
In the corresponding song features information of pre-stored each song, each song in the first concern list of songs is determined
Bent corresponding song features information;
According to the corresponding song features information of each song in the first concern list of songs, first account pair is determined
The account features information answered.
13. device according to claim 12, which is characterized in that first determining module is used for:
Determine concern type of first account to each song in the first concern list of songs;And determine described the
One account pays close attention to concern duration of the moment apart from current time to each song in the first concern list of songs;
According to the pre-stored different correspondences paid close attention to type and pay close attention to weights, determine first account to described first
Pay close attention to the corresponding concern weights of concern type of each song in list of songs;And according to time decay formula and described
First account pays close attention to concern duration of the moment apart from current time to each song in the first concern list of songs, really
Determine the corresponding time dough softening of each song;
By multiplying for corresponding concern weights and each corresponding time dough softening of song of the concern type of each song
Product calculates each song in the first concern list of songs as the weights of the corresponding song features information of each song
The weighted sum of corresponding song features information obtains the corresponding account features information of first account.
14. device according to claim 10, which is characterized in that second determining module is used for:
It singly concentrates each song single candidate's song, in the corresponding song list characteristic information of pre-stored each song list, determines institute
State the single characteristic information of the single corresponding song of song;
It is concerned information according to song list is corresponding, determines the single attribute information of the single corresponding song of the song;
By the single single attribute information of corresponding song of the song and the single characteristic information of the single corresponding song of the song, it is determined as the song single pair
The song list information answered.
15. the device according to claim 12 or 14, which is characterized in that described device further includes:
4th determining module, the identification information for determining each song in library;By each song in the library
Bent identification information is input in the song features extraction model of training gone out in advance, determines the song of each song in library
Bent characteristic information;It is single for pre-stored each song, determine the corresponding song features information of each song that the song includes singly;
The corresponding song features information of each song for including singly is sung according to described, determines the single characteristic information of the single corresponding song of the song.
16. device according to claim 15, which is characterized in that described device further includes:
Memory module, for for it is pre-stored it is each sing include singly each song, obtain the song identification information,
And the identification information of the first preset number song before the song and the song subsequent second preset number
The identification information of song, respectively using the identification information of the song as mode input data, by first preset number
As model output data, training set is arrived in storage for the identification information of song and the identification information of the second preset number song
In each mode input data and model output data correspondence in;
First training module, for the correspondence according to each mode input data and model output data in training set, base
In the model output data obtained by the song features extraction model comprising undetermined parameter levels off to the correspondence
The training philosophy of model output data is trained the song features extraction model comprising undetermined parameter, obtains song features
Extraction model.
17. device according to claim 10, which is characterized in that described device further includes:
Second training module, for according to each account in the training set scoring and each account corresponding account single to each song
Information respectively sings the single information of single corresponding song, based on commenting for the account singing in antiphonal style list predicted by the Rating Model comprising undetermined parameter
The training philosophy for dividing the single scoring of account singing in antiphonal style in leveling off to training set, instructs the Rating Model comprising undetermined parameter
Practice, obtains Rating Model.
18. the device according to claim 10 or 16, which is characterized in that described device further includes:
Computing module, for for each account, obtaining the corresponding concern list of songs of the account;It is corresponded to according to the account
Concern list of songs, count the account it is corresponding concern list of songs in different genre types song number, and
The number of the song of the different genre types in the corresponding concern list of songs of the account is calculated in the corresponding pass of the account
The accounting in the total number of the song in list of songs is noted, the song accounting of the corresponding different genre types of the account is obtained;
Module is chosen, in each account, the song accounting for choosing corresponding default genre type to reach default accounting threshold value
Target account;
5th determining module, for determining the corresponding account information of target account;
Second prediction module, described in for each song in the corresponding concern list of songs of the target account, determining
The song information of the account information of the target account and the song is input to and is trained in advance by the song information of song
Rating Model, predict scoring of the target account to the song;
6th determining module, in each song in the corresponding concern list of songs of the target account, determining corresponding
Score maximum third preset number song;
Creation module, for creating the song list being made of the third preset number song.
19. a kind of server, which is characterized in that the server includes processor and memory, is stored in the memory
At least one instruction, at least one section of program, code set or instruction set, at least one instruction, at least one section of program, the institute
Code set or instruction set is stated to be loaded by the processor and executed to realize that it is single that the transmission as described in claim 1 to 9 is any is sung
The method of mark.
20. a kind of computer readable storage medium, which is characterized in that be stored at least one instruction, extremely in the storage medium
Few one section of program, code set or instruction set, at least one instruction, at least one section of program, the code set or the instruction
Collection is loaded by processor and is executed to realize that the method singly identified is sung in the transmission as described in claim 1 to 9 is any.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810169174.9A CN108388652B (en) | 2018-02-28 | 2018-02-28 | Method and device for sending song list identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810169174.9A CN108388652B (en) | 2018-02-28 | 2018-02-28 | Method and device for sending song list identification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108388652A true CN108388652A (en) | 2018-08-10 |
CN108388652B CN108388652B (en) | 2021-01-12 |
Family
ID=63069571
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810169174.9A Active CN108388652B (en) | 2018-02-28 | 2018-02-28 | Method and device for sending song list identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108388652B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615731A (en) * | 2018-11-10 | 2019-04-12 | 东莞市华睿电子科技有限公司 | A kind of audio-frequency inputting method based on recognition of face |
CN111078931A (en) * | 2019-12-10 | 2020-04-28 | 腾讯科技(深圳)有限公司 | Singing sheet pushing method and device, computer equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120078824A1 (en) * | 2002-10-03 | 2012-03-29 | Antonio Trias Bonet | Method and System for Music Recommendation Based on Immunology |
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 |
CN106610968A (en) * | 2015-10-21 | 2017-05-03 | 广州酷狗计算机科技有限公司 | Song menu list determination method and apparatus, and electronic device |
CN106649686A (en) * | 2016-12-16 | 2017-05-10 | 天翼爱音乐文化科技有限公司 | User interest group dividing method and system based on multilayer potential features |
CN107577795A (en) * | 2017-09-19 | 2018-01-12 | 广州酷狗计算机科技有限公司 | A kind of methods, devices and systems for recommending song |
-
2018
- 2018-02-28 CN CN201810169174.9A patent/CN108388652B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120078824A1 (en) * | 2002-10-03 | 2012-03-29 | Antonio Trias Bonet | Method and System for Music Recommendation Based on Immunology |
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 |
CN106610968A (en) * | 2015-10-21 | 2017-05-03 | 广州酷狗计算机科技有限公司 | Song menu list determination method and apparatus, and electronic device |
CN106649686A (en) * | 2016-12-16 | 2017-05-10 | 天翼爱音乐文化科技有限公司 | User interest group dividing method and system based on multilayer potential features |
CN107577795A (en) * | 2017-09-19 | 2018-01-12 | 广州酷狗计算机科技有限公司 | A kind of methods, devices and systems for recommending song |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615731A (en) * | 2018-11-10 | 2019-04-12 | 东莞市华睿电子科技有限公司 | A kind of audio-frequency inputting method based on recognition of face |
CN111078931A (en) * | 2019-12-10 | 2020-04-28 | 腾讯科技(深圳)有限公司 | Singing sheet pushing method and device, computer equipment and storage medium |
CN111078931B (en) * | 2019-12-10 | 2023-08-01 | 腾讯科技(深圳)有限公司 | Song list pushing method, device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108388652B (en) | 2021-01-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108874823A (en) | The implementation method and device of intelligent customer service | |
CN108764994A (en) | A kind of user behavior guidance method, device, server and storage medium | |
CN106156092B (en) | Data processing method and device | |
CN108509617A (en) | Construction of knowledge base, intelligent answer method and device, storage medium, the terminal in knowledge based library | |
JP2017220238A (en) | Method and device for providing answer in question answering system | |
WO2020257991A1 (en) | User identification method and related product | |
CN111523035B (en) | Recommendation method, device, server and medium for APP browsing content | |
CN109086317A (en) | Risk control method and relevant apparatus | |
CN107093091A (en) | A kind of data processing method and device | |
CN108401005B (en) | Expression recommendation method and device | |
CN111061948A (en) | User label recommendation method and device, computer equipment and storage medium | |
CN111428885B (en) | User indexing method in federated learning and federated learning device | |
CN108388652A (en) | A kind of method and apparatus for sending song and singly identifying | |
CN107038165B (en) | Service parameter acquisition method and device | |
CN108536787A (en) | content identification method and device | |
CN111309994A (en) | User matching method and device, electronic equipment and readable storage medium | |
CN116860953B (en) | Question-answer matching method and system based on question-answer system | |
CN112053245B (en) | Information evaluation method and system | |
CN117131401A (en) | Object recognition method, device, electronic equipment and storage medium | |
CN109697224A (en) | A kind of bill message treatment method, device and storage medium | |
CN106897208A (en) | A kind of quantization method and device of application software Consumer's Experience | |
CN106844504A (en) | A kind of method and apparatus for sending the single mark of song | |
CN109255016A (en) | Answer method, device and computer readable storage medium based on deep learning | |
CN114386688B (en) | User intention prediction method and system based on multi-data fusion | |
CN115186063A (en) | Service data processing method, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | 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 |