CN108255840A - A kind of recommendation method and system of song - Google Patents
A kind of recommendation method and system of song Download PDFInfo
- Publication number
- CN108255840A CN108255840A CN201611239007.4A CN201611239007A CN108255840A CN 108255840 A CN108255840 A CN 108255840A CN 201611239007 A CN201611239007 A CN 201611239007A CN 108255840 A CN108255840 A CN 108255840A
- Authority
- CN
- China
- Prior art keywords
- song
- user
- audio
- scoring
- similar
- 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
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/683—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of recommendation method and system of song, the method includes:Song is listened behavioural information and to listen song source-information according to user, obtain corresponding user and listen song preference pattern;Based on pre-set rule, corresponding song audio feature relevance model is obtained;The song and the current song for selection for listening song preference pattern selection user preference according to user, the similar song of corresponding audio is got, and the song similar to audio scores to be scored accordingly according to song audio feature relevance model;The scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended.Song audio feature relevance model in the embodiment of the present invention, the similar songs database of song recommendations can be carried out by enriching, solve the problems, such as that the number of songs of current similar songs database is few, further, similitude based on song on the audio frequency characteristics such as beat, tone, tone color carries out song recommendations, improves the accuracy of song recommendations.
Description
Technical field
The present invention relates to Audiotechnica field, specifically, the recommendation method and system the present invention relates to a kind of song.
Background technology
At present, the recommendation method of common song often there are two types of, a kind of is that the song of the collaborative filtering based on article pushes away
Method is recommended, another is the song recommendations method of the content based on song.
First, the principle of the song recommendations method of the collaborative filtering based on article be used to differentiating current certain it is specific
Customer may interested song to carry out song recommendations.And it is based on it that current certain specific consumers are recommended with the conclusion of song
His similar customers are to the interested analysis of which song.
The song recommendations method of collaborative filtering based on article, that analyzes user listens song interest, and spy is found in user group
Determine the user of the similar interests of user, it is specified to this to form system for evaluation of these the comprehensive similar users to a certain song information
Prediction of the user to the fancy grade of this song information.
In practical applications, the song recommendations method of the common collaborative filtering based on article is specially the association based on item
With filtering, the similitude between item is predicted to the scoring of different item by user, is done based on the similitude between item
Go out to recommend.
One of the shortcomings that song recommendations method of collaborative filtering based on article, is, due to evaluation of the user to same song
Information is very sparse, and therefore, the accuracy that the song based on user evaluates the similitude between obtained user is just very low.It is based on
The shortcomings that song recommendations method of the collaborative filtering of article second is that, increasing and the song recommended needed to get over user
More, the song recommendations method of the collaborative filtering based on article obtains active user's efficiency that really interested song is recommended
It is lower.The three of the shortcomings that song recommendations method of collaborative filtering based on article are, even if user's first is delithted with current song,
But user then likes active user's second of same song to be impossible to there is no being evaluated current song with user's first
Obtain the recommendation of user's first.Therefore, the practicability of the song recommendations method of the collaborative filtering based on article is not high.
Secondly, the song recommendations method of the content based on song is common song recommendations method.By analyzing song
Content, so as to have the song recommendations of Similar content to similar user.The song recommendations method of content based on song
Shortcoming is that the information difficult to realize analyzed based on song content carries out machine automatic fitration.In this way so that recommend the phase of user
It is often few like song, and recommend the renewal speed of song very slow.The real-time of song recommendations is very poor, reduces user's body
It tests.
Invention content
The embodiment of the present invention is to provide a kind of recommendation method and system of song, passes through song audio feature degree of correlation mould
Type enriches the similar songs database that can carry out song recommendations, the number of songs for solving current similar songs database
The problem of few, further, the similitude based on song on the audio frequency characteristics such as beat, tone, tone color, carry out song recommendations, carry
The high accuracy of song recommendations.
In a first aspect, an embodiment of the present invention provides a kind of recommendation method of song, the method includes:
Song is listened behavioural information and to listen song source-information according to user, obtain corresponding user and listen song preference pattern;
Based on pre-set rule, corresponding song audio feature relevance model is obtained;
The song and the current song for selection that song preference pattern selection user preference is listened according to the user, root
The similar song of corresponding audio is got, and the similar song of audio is carried out according to the song audio feature relevance model
It scores to be scored accordingly;
The scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended.
Preferably, it is described based on pre-set rule, it obtains corresponding song audio feature relevance model and specifically wraps
It includes:
Based on the audio frequency characteristics of marsyas source codes extraction current song, corresponding audio data is obtained;
Root according to the rule pre-set, is standardized corresponding audio data, the song after being standardized accordingly
Bent audio data;
Based on COS distance, COS distance is calculated to the document vector of the song after various criterion, obtains corresponding phase
Guan Du, and the obtained degree of correlation is ranked up, the preferential selected as similar songs of the forward song that sorts;
Summarize user's relevance data, obtain corresponding song audio feature relevance model.
Preferably, the method further includes:
The audio frequency characteristics of current song are obtained, wherein, the audio frequency characteristics specifically include following at least one audio frequency characteristics:
Mel frequency cepstral coefficients, tone color, zero phase transformation and beat.
Preferably, the calculation formula that is standardized to corresponding audio data is specially:
Wherein, n is any positive integer from 1 to N.
Preferably, the method further includes:Calculate the scoring of the song similar to current song audio, wherein, calculate and
Calculation formula is specially used by the scoring of the similar song of current song audio:
scorej=∑i∈sSim (i, j) * pref (i), wherein, scorejRecommend the score of song j in song, S for user
For the song of user's history preference, sim (i, j) is the audio similarity of song i and song j, and pref (i) is user to song i
Preference degree.
Preferably, it is described song to be listened behavioural information and to listen song source-information according to user, it obtains corresponding user and listens song
Preference pattern specifically includes:
The daily record that the user in the predetermined time plays song is extracted, listens song behavioural information to get corresponding user;
Parsing user listens the user in song behavioural information to listen the old song form to be, and listens the weighted value and listen song that an old song form is according to user
The weighted value in source counts the initial scoring that the user in the predetermined time listens song;
Based on time decay calculation formula, the initial scoring of the listened song is handled, obtains the corresponding time
Song scoring after attenuation;
Song amount is broadcast based on song, the song scoring after decaying to the time carries out temperature drop power, is sung accordingly
Bent preference;
Summarize corresponding song preference degrees of data, and song preference data are modified, to obtain corresponding user
Listen song preference pattern.
Preferably, calculating user listens the formula initially to score of song to be specially:
Score=listen*weight, wherein, score is listened the initial scoring of song by user, according to listen
Judge whether song completely listens to obtained score, weight is the different different weights listened corresponding to the source of song
Value.
Preferably, obtaining the time decay calculation formula that the song after corresponding time attenuation scores is specially:
Score=Snow+Shistory* decay_factor, wherein, score is to carry out time attenuation to the initial scoring
Handle the song scoring after obtained time attenuation, SnowThe scoring of song, S are listened for test same day userhistoryTo be listened
The history scoring of song, decay_factor is decay factor.
Preferably, the calculation formula of calculating song preference is specially:
Wherein, preference is song preference
Degree, score are the song scoring carried out to the initial scoring after the obtained time attenuation of time attenuation processing, and CNT is song
Bent to broadcast song amount, A is song temperature, A log20(CNT+20)。
Second aspect, an embodiment of the present invention provides a kind of commending system of song, the system comprises:
Song preference pattern acquiring unit is listened, song is listened behavioural information and to listen song source-information according to user, obtained corresponding
User listen song preference pattern;
Relevance model acquiring unit based on pre-set rule, obtains corresponding song audio feature degree of correlation mould
Type;
Audio similar songs obtain and scoring unit, and the song of song preference pattern selection user preference is listened according to the user
It is similar to get corresponding audio according to the song audio feature relevance model for song and the current song for selection
Song, and the song similar to audio scores to be scored accordingly;
Marking and queuing and recommendation unit, the scoring of the song similar to audio are ranked up, and the high similar songs that score are excellent
First recommended.
An embodiment of the present invention provides a kind of recommendation method and system of song, wherein, the method includes:According to user
Song is listened behavioural information and to listen song source-information, obtain corresponding user and listen song preference pattern;Based on pre-set rule,
Obtain corresponding song audio feature relevance model;According to user listen song preference pattern choose user preference song and
For the current song of selection, the similar song of corresponding audio is got, and right according to song audio feature relevance model
The similar song of audio scores to be scored accordingly;The scoring of the song similar to audio is ranked up, and scoring is high
Similar songs preferentially recommended.Song audio feature relevance model in the embodiment of the present invention, enriching can carry out
The similar songs database of song recommendations solves the problems, such as that the number of songs of current similar songs database is few, further,
Similitude based on song on the audio frequency characteristics such as beat, tone, tone color carries out song recommendations, improves the accurate of song recommendations
Property.
Description of the drawings
Fig. 1 is a kind of flow chart of the recommendation method of song provided in an embodiment of the present invention;
Fig. 2 is a kind of structure diagram of the commending system of song provided in an embodiment of the present invention.
Specific embodiment
Purpose, technical scheme and advantage to make the embodiment of the present invention are clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical solution in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
Part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
All other embodiments obtained without making creative work shall fall within the protection scope of the present invention.
For ease of the understanding to the embodiment of the present invention, it is further explained below in conjunction with attached drawing with specific embodiment
It is bright.
In technical solution provided by the present invention, song is listened behavioural information and to listen song source-information according to user, obtained
Corresponding user listens song preference pattern;Based on pre-set rule, corresponding song audio feature relevance model is obtained;Root
The song and the current song for selection for listening song preference pattern selection user preference according to user, according to song audio feature
Relevance model gets the similar song of corresponding audio, and the song similar to audio scores to be commented accordingly
Point;The scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended.As a result of song
Audio frequency characteristics relevance model, the song audio feature relevance model are in the audios such as beat, tone, tone color spy based on song
Similitude in sign and the song recommendations model established.Song recommendations are carried out by the song audio feature relevance model, no
The similar songs database of song recommendations can be carried out by only enriching, and the number of songs for solving current similar songs database is few
The problem of.Further, the accuracy of song recommendations is also improved so that the song for recommending user is more in line with the reality of user
Border interest, improves user experience.
The technical solution that the invention will now be described in detail with reference to the accompanying drawings.
Fig. 1 is a kind of flow chart of the recommendation method of song provided in an embodiment of the present invention, as shown in Figure 1, a kind of song
Recommendation method include the following steps:
S101:Song is listened behavioural information and to listen song source-information according to user, obtain corresponding user and listen song preference mould
Type.
Specifically, song is listened behavioural information and to listen song source-information according to user, obtain corresponding user and listen song inclined
Good model specifically includes:
The daily record that the user in the predetermined time plays song is extracted, listens song behavioural information to get corresponding user;
Parsing user listens the user in song behavioural information to listen the old song form to be, and listens the weighted value and listen song that an old song form is according to user
The weighted value in source counts the initial scoring that the user in the predetermined time listens song;
Based on time decay calculation formula, the initial scoring of listened song is handled, obtains corresponding time attenuation
Song scoring afterwards;
Song amount is broadcast based on song, the song scoring after decaying to the time carries out temperature drop power, and it is inclined to obtain corresponding song
Good degree;
Summarize corresponding song preference degrees of data, and song preference data are modified, to obtain corresponding user
Listen song preference pattern.
In order to which the dynamic for reflecting user listens an old song form to be and listen the migration of song preference, user listens song preference pattern to can be used for point
The old song form of listening for analysing the short-term preference of user is that can be used for the old song form of listening of long-term preference of analysis user to be.Wherein, user
An old song form of listening for short-term preference is to refer to listen song preference behavior within user seven days, and an old song form of listening for the long-term preference of user is yes
Refer to and listen song preference behavior within user six months.
In specific practical application, user listen song preference pattern to establish process as described below:
First, parsing music box broadcasts song daily record, song data is listened the user got to analyze according to scheduled rule,
The counting user same day listens song preference, to listen song data count, to user listen an old song form be and listen song source assign not
Same weight.
Calculating user listens the formula initially to score of song to be specially:
Score=listen*weight, wherein, score is listened the initial scoring of song by user, according to listen
Judge whether song completely listens to obtained score, weight is the different different weights listened corresponding to the source of song
Value.
When user listen completely listens to song, the numerical value of listen is chosen for 1.In practical applications, in order to more preferable
Ground defines whether user completely listens to song, and the situation that song is completely listened to user is quantified.Specifically, user is complete
Whole song of listening to refers to that user's listens song duration to be more than or equal to the 40% of song total duration., when the duration that user listens to song is small
0 is chosen in the numerical value of 40%, listen of song total duration.
Weight is the different different weighted values listened corresponding to the source of song.Specifically, when song origin is red
During heart list, then corresponding weighted value is 80;When song origin is downloads, then corresponding weighted value is 50;Work as song origin
During for search, then corresponding weighted value is 30;When song origin is local, then corresponding weighted value is 20;Work as song origin
During for self-built list, then corresponding weighted value is 20;When song origin is the default list, then corresponding weighted value is 20;When
When song origin is other, then corresponding weighted value is 1;When song origin is radio station, then corresponding weighted value is 0.2;When
When song origin is downloaded to cancel, then corresponding weighted value is -10.
It is this to broadcast song daily record by parsing music box, based on above-mentioned rule, big data is counted, it is possible to know use
The song preference on the family same day.
And then the old song form of listening for merging the user in certain time is, if to get user listens song preference pattern in short term,
Then merging the old song form of listening within nearest 3 days of user is.Song preference pattern is listened for a long time to get user, then merges user 6
Month in an old song form of listening be.Further, the initial scoring of song is listened user to carry out the calculating of time attenuation, temperature is carried out and declines
Subtract and normalize and temperature drop power calculating, to update preference of the user to song.
Specifically, the time decay calculation formula for obtaining the song scoring after corresponding time attenuation is specially:Score
=Snow+Shistory* decay_factor, wherein, score is to carry out the time attenuation processing obtained time to initially scoring
Song scoring after attenuation, SnowThe scoring of song is listened for test same day user, if active user does not listen to song, Snow
It is denoted as 0.ShistoryHistory for listened song scores, and decay_factor is decay factor.In a particular embodiment, decay_
The numerical value that factor chooses is 0.8.
Further, to the song of elapsed time attenuation processing, temperature attenuation and normalized are carried out.
Specifically, the song amount of broadcasting based on listened song makees temperature drop power.
Calculate song preference calculation formula be specially:
Wherein, preference is song preference
Degree, score are to score the song after carrying out the obtained time attenuation of time attenuation processing that initially scores, and CNT is song
Broadcast song amount, A is song temperature, A log20(CNT+20)。
S102:Based on pre-set rule, corresponding song audio feature relevance model is obtained.
In one particular embodiment of the present invention, based on pre-set rule, corresponding song audio feature is obtained
Relevance model specifically includes:
Based on the audio frequency characteristics of marsyas source codes extraction current song, corresponding audio data is obtained.Need what is illustrated
It is that marsyas is the source code for carrying out audio analysis, can identify the music semantic feature such as music rhythm, musical instrument, use hidden Ma Er
It can husband's model.
Further, the audio frequency characteristics of the current song based on the extraction of marsyas source codes specifically include following at least one
Audio frequency characteristics:Mel frequency cepstral coefficients, tone color, zero phase transformation and beat.In practical application scene, based on marsyas source codes
The more than aforementioned four audio frequency characteristics of audio frequency characteristics of the current song of extraction, one shares 124 audio frequency characteristics, herein no longer one by one
It repeats.
Root according to the rule pre-set, is standardized corresponding audio data, the song after being standardized accordingly
Bent audio data.
It should be noted that in order to improve the accuracy of the song audio feature relevance model of foundation, to corresponding sound
Frequency evidence is standardized.Wherein, the calculation formula being standardized to corresponding audio data is specially:Wherein, n is any positive integer from 1 to N.
Based on COS distance, COS distance is calculated to the document vector of the song after various criterion, obtains corresponding phase
Guan Du, and the obtained degree of correlation is ranked up, the preferential selected as similar songs of the forward song that sorts.
In one particular embodiment of the present invention, based on COS distance, to the document of the song after various criterion to
Amount calculates COS distance, obtains the corresponding degree of correlation and specifically includes following steps:
Two of COS distance to be calculated are extracted in the document vector of song after the standardization listened to from different user
The document vector of song after the standardization that user listens to;
Based on COS distance, the cosine value of the document vector of the song after the standardization that two users listen to is calculated;
If the document vectorial angle cosine value of the song after the standardization that two users listen to obtains two close to 1
The song degree of correlation that a user listens to is high.Therefore, song recommendations can be carried out mutually between two users.If conversely, two use
The document vectorial angle cosine value of song after the standardization that family is listened to is smaller, then it is related to obtain the song that two users listen to
Degree is lower.Therefore, it is not recommended that carry out song recommendations between two users.
Summarize user's relevance data, obtain corresponding song audio feature relevance model.
S103:The song and the current song for selection that song preference pattern selection user preference is listened according to user, root
The similar song of corresponding audio is got, and the song similar to audio scores according to song audio feature relevance model
To be scored accordingly.
Further, before the song similar to audio scores, the song similar to current song audio is calculated
Scoring, wherein, calculation formula is specially used by calculating the scoring of the song similar to current song audio:scorej=
∑i∈sSim (i, j) * pref (i), wherein, scorejRecommend the score of song j in song for user, S is user's history preference
Song, sim (i, j) are the audio similarity of song i and song j, and pref (i) is preference degree of the user to song i.
S104:The scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended.It needs
Illustrate, before the scoring of the song similar to audio is ranked up, need to filter out the song that active user listened.
In practical applications, in order to improve the efficiency of the recommendation method of song provided by the present invention, real-time is embodied,
The obtained degree of correlation is ranked up, 200 song selected as similar songs before relevancy ranking.
In specific practical application, the process for recommending song is described in detail below:
Firstly, for the user extract relevancy ranking before 200 song as similar songs.
Secondly, the song of short-term preference, the song of long-term preference are taken respectively to user.Wherein, the song of short-term preference is
Refer to the song listened within nearest seven days, and the song of long-term preference refers to the song listened within 6 months.
Then, to each song in preference, the similar song of audio is found, is calculated similar to current song audio
The scoring of song, and the score of accumulative similar songs.Wherein, the scoring for calculating the song similar to current song audio is used
Calculation formula be specially:scorej=∑i∈sSim (i, j) * pref (i), wherein, scorejRecommend song in song for user
The score of j, S are the song of user's history preference, and sim (i, j) is the audio similarity of song i and song j, and pref (i) is uses
Family is to the preference degree of song i.
Then, the song that filtering user listened;
Finally, the scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended, and will
The song recommendations that user did not listen are to user.
In conclusion a kind of recommendation method of song provided in an embodiment of the present invention, listens song behavioural information according to user
And song source-information is listened, it obtains corresponding user and listens song preference pattern;Based on pre-set rule, corresponding song is obtained
Audio frequency characteristics relevance model;Song preference pattern is listened to choose the song of user preference and for the current of selection according to user
Song gets the similar song of corresponding audio, and the song similar to audio according to song audio feature relevance model
It scores to be scored accordingly;The scoring of the song similar to audio is ranked up, and the high similar songs that score are preferential
Recommended.Technical solution provided by the present invention carries out song recommendations by song audio feature relevance model, not only rich
The rich similar songs database that can carry out song recommendations, the number of songs for solving current similar songs database few are asked
Topic.Further, the accuracy of song recommendations is also improved.According to statistics, it is sung by technical solution provided by the present invention
Song is recommended, and relative to current technology, after the song for recommending user, user is complete, and audience rating improves 10%.In addition, by
The actual interest of user is more in line in the song that user is recommended by song audio feature relevance model, according to statistics, is used
The ratio that the participation at family is listened to improves 15% compared to current technology, thus, it can be known that greatly increasing user experience.
As shown in Fig. 2, the commending system of a kind of song that the embodiment of the present invention is provided, including:Song preference pattern is listened to obtain
Take unit 201, relevance model acquiring unit 202, the acquisition of audio similar songs and scoring unit 203 and marking and queuing and recommendation
Unit 204.
Specifically, listen song preference pattern acquiring unit, listening song behavioural information and listen song source-information according to user,
It obtains corresponding user and listens song preference pattern;
Relevance model acquiring unit based on pre-set rule, obtains corresponding song audio feature degree of correlation mould
Type;
Audio similar songs obtain and scoring unit, and the song of song preference pattern selection user preference is listened according to user, with
And the current song for selection, the similar song of corresponding audio is got according to song audio feature relevance model, and
The song similar to audio scores to be scored accordingly;
Marking and queuing and recommendation unit, the scoring of the song similar to audio are ranked up, and the high similar songs that score are excellent
First recommended.
Further, relevance model acquiring unit is specifically used for:Audio based on marsyas source codes extraction current song
Feature obtains corresponding audio data;
Root according to the rule pre-set, is standardized corresponding audio data, the song after being standardized accordingly
Bent audio data;
Based on COS distance, COS distance is calculated to the document vector of the song after various criterion, obtains corresponding phase
Guan Du, and the obtained degree of correlation is ranked up, the preferential selected as similar songs of the forward song that sorts;
Summarize user's relevance data, obtain corresponding song audio feature relevance model.
Further, the commending system of a kind of song that the embodiment of the present invention is provided further includes acquiring unit (in fig. 2
It does not mark).
Specifically, acquiring unit obtains the audio frequency characteristics of current song, wherein, the audio frequency characteristics tool that acquiring unit obtains
Body includes following at least one audio frequency characteristics:Mel frequency cepstral coefficients, tone color, zero phase transformation and beat.
Further, the calculation formula being standardized in relevance model acquiring unit to corresponding audio data is specific
For:Wherein, n is any positive integer from 1 to N.
Further, song preference pattern acquiring unit is listened to be specifically used for:
The daily record that the user in the predetermined time plays song is extracted, listens song behavioural information to get corresponding user;
Parsing user listens the user in song behavioural information to listen the old song form to be, and listens the weighted value and listen song that an old song form is according to user
The weighted value in source counts the initial scoring that the user in the predetermined time listens song;
Based on time decay calculation formula, the initial scoring of listened song is handled, obtains corresponding time attenuation
Song scoring afterwards;
Song amount is broadcast based on song, the song scoring after decaying to the time carries out temperature drop power, and it is inclined to obtain corresponding song
Good degree;
Summarize corresponding song preference degrees of data, and song preference data are modified, to obtain corresponding user
Listen song preference pattern.
Wherein, listen in song preference pattern acquiring unit listens the formula initially to score of song specific for calculating user
For:
Score=listen*weight, wherein, score is listened the initial scoring of song by user, according to listen
Judge whether song completely listens to obtained score, weight is the different different weights listened corresponding to the source of song
Value.
Wherein, listen in song preference pattern acquiring unit for be calculated the song scoring after the attenuation of corresponding time when
Between decay calculation formula be specially:
Score=Snow+Shistory* decay_factor, wherein, score is to carry out time attenuation processing to initially scoring
Song scoring after obtained time attenuation, SnowThe scoring of song, S are listened for test same day userhistoryTo be listened song
History scoring, decay_factor is decay factor.
Wherein, the calculation formula that song preference pattern acquiring unit is used to calculate song preference is listened to be specially:
Wherein, preference is song preference
Degree, score are to score the song after carrying out the obtained time attenuation of time attenuation processing that initially scores, and CNT is song
Broadcast song amount, A is song temperature, A log20(CNT+20)。
Further, audio similar songs obtain and scoring unit is specifically used for calculating the song similar to current song audio
Bent scoring, wherein, the scoring institute for the song that audio similar songs obtain and scoring unit calculating is similar to current song audio
The calculation formula of use is specially:
scorej=∑i∈sSim (i, j) * pref (i), wherein, scorejRecommend the score of song j in song, S for user
For the song of user's history preference, sim (i, j) is the audio similarity of song i and song j, and pref (i) is user to song i
Preference degree.
In technical scheme of the present invention, song is listened behavioural information and to listen song source-information according to user, obtained corresponding
User listens song preference pattern;Based on pre-set rule, corresponding song audio feature relevance model is obtained;According to user
The song and the current song for selection that song preference pattern chooses user preference are listened, according to the song audio feature degree of correlation
Model gets the similar song of corresponding audio, and the song similar to audio scores to be scored accordingly;It is right
The scoring of the similar song of audio is ranked up, and the high similar songs that score preferentially are recommended.Technology provided by the present invention
Scheme carries out song recommendations by song audio feature relevance model, and the similar of song recommendations can be carried out by not only enriching
Song database solves the problems, such as that the number of songs of current similar songs database is few.Further, song is also improved to push away
The accuracy recommended.Further, since the song that user is recommended by song audio feature relevance model is more in line with user's
Actual interest greatly increases user experience.
Above-described specific embodiment has carried out the purpose of the present invention, technical solution and advantageous effect further
It is described in detail, it should be understood that the foregoing is merely the specific embodiment of the present invention, is not intended to limit the present invention
Protection domain, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of recommendation method of song, which is characterized in that including:
Song is listened behavioural information and to listen song source-information according to user, obtain corresponding user and listen song preference pattern;
Based on pre-set rule, corresponding song audio feature relevance model is obtained;
The song and the current song for selection that song preference pattern selection user preference is listened according to the user, according to institute
It states song audio feature relevance model and gets the similar song of corresponding audio, and the song similar to audio scores
To be scored accordingly;
The scoring of the song similar to audio is ranked up, and the high similar songs that score preferentially are recommended.
2. according to the method described in claim 1, it is characterized in that, described based on pre-set rule, sung accordingly
Bent audio frequency characteristics relevance model specifically includes:
Based on the audio frequency characteristics of marsyas source codes extraction current song, corresponding audio data is obtained;
Root according to the rule pre-set, is standardized corresponding audio data, the song after being standardized accordingly
Audio data;
Based on COS distance, COS distance is calculated to the document vector of the song after various criterion, obtains the corresponding degree of correlation,
And the obtained degree of correlation is ranked up, the preferential selected as similar songs of the forward song that sorts;
Summarize user's relevance data, obtain corresponding song audio feature relevance model.
3. according to the method described in claim 1, it is characterized in that, the method further includes:
The audio frequency characteristics of current song are obtained, wherein, the audio frequency characteristics specifically include following at least one audio frequency characteristics:
Mel frequency cepstral coefficients, tone color, zero phase transformation and beat.
4. the according to the method described in claim 1, it is characterized in that, calculating being standardized to corresponding audio data
Formula is specially:
Wherein, n is any positive integer from 1 to N.
5. according to the method described in claim 1, it is characterized in that, the method further includes:It calculates and current song audio phase
As song scoring, wherein, calculation formula is specially used by calculating the scoring of the song similar to current song audio:
scorej=∑i∈sSim (i, j) * pref (i), wherein, scorejRecommend the score of song j in song for user, S is uses
The song of family history preference, sim (i, j) are the audio similarity of song i and song j, and pref (i) is happiness of the user to song i
Good degree.
6. according to the method described in claim 1, it is characterized in that, it is described according to user listen song behavioural information and listen song come
Source information obtains corresponding user and song preference pattern is listened to specifically include:
The daily record that the user in the predetermined time plays song is extracted, listens song behavioural information to get corresponding user;
Parsing user listens the user in song behavioural information to listen the old song form to be, and listens the weighted value and listen song source that an old song form is according to user
Weighted value, count the initial scoring that the user in the predetermined time listens song;
Based on time decay calculation formula, the initial scoring of the listened song is handled, obtains corresponding time attenuation
Song scoring afterwards;
Song amount is broadcast based on song, the song scoring after decaying to the time carries out temperature drop power, and it is inclined to obtain corresponding song
Good degree;
Summarize corresponding song preference degrees of data, and song preference data are modified, listen song to obtain corresponding user
Preference pattern.
7. according to the method described in claim 6, listen the formula initially to score of song specific it is characterized in that, calculating user
For:
Score=listen*weight, wherein, score is listened the initial scoring of song by user, and listen is according to judgement
Whether song completely listens to obtained score, and weight is the different different weighted values listened corresponding to the source of song.
8. according to the method described in claim 6, it is characterized in that, obtain the time of the song scoring after corresponding time attenuation
Decay calculation formula is specially:
Score=Snow+Shistory* decay_factor, wherein, score is to carry out time attenuation processing to the initial scoring
Song scoring after obtained time attenuation, SnowThe scoring of song, S are listened for test same day userhistoryTo be listened song
History scoring, decay_factor is decay factor.
9. according to the method described in claim 6, it is characterized in that, the calculation formula for calculating song preference is specially:
Wherein, preference is song preference,
Score is the song scoring carried out to the initial scoring after the obtained time attenuation of time attenuation processing, and CNT is song
Broadcast song amount, A is song temperature, A log20(CNT+20)。
10. a kind of commending system of song, which is characterized in that including:
Song preference pattern acquiring unit is listened, song is listened behavioural information and to listen song source-information according to user, is used accordingly
Listen song preference pattern in family;
Relevance model acquiring unit based on pre-set rule, obtains corresponding song audio feature relevance model;
Audio similar songs obtain and scoring unit, and the song of song preference pattern selection user preference is listened according to the user, with
And the current song for selection, the similar song of corresponding audio is got according to the song audio feature relevance model
Song, and the song similar to audio scores to be scored accordingly;
Marking and queuing and recommendation unit, the scoring of the song similar to audio are ranked up, score high similar songs preferentially into
Row is recommended.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611239007.4A CN108255840B (en) | 2016-12-28 | 2016-12-28 | Song recommendation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611239007.4A CN108255840B (en) | 2016-12-28 | 2016-12-28 | Song recommendation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108255840A true CN108255840A (en) | 2018-07-06 |
CN108255840B CN108255840B (en) | 2021-04-02 |
Family
ID=62720423
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611239007.4A Active CN108255840B (en) | 2016-12-28 | 2016-12-28 | Song recommendation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108255840B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920689A (en) * | 2018-07-12 | 2018-11-30 | 江苏满运软件科技有限公司 | Source of goods recommended method and system |
CN109543132A (en) * | 2018-11-22 | 2019-03-29 | 深圳墨世科技有限公司 | Content recommendation method, device, electronic equipment and storage medium |
CN109582817A (en) * | 2018-10-30 | 2019-04-05 | 努比亚技术有限公司 | A kind of song recommendations method, terminal and computer readable storage medium |
CN110008369A (en) * | 2018-12-26 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Information processing method and its device, electronic equipment, computer-readable medium |
CN110457492A (en) * | 2019-06-06 | 2019-11-15 | 北京酷我科技有限公司 | A kind of individuation data proposed algorithm and recommender system |
CN110909199A (en) * | 2019-11-18 | 2020-03-24 | 彩讯科技股份有限公司 | Song recommendation display method and device based on vehicle-mounted system |
CN111915409A (en) * | 2020-08-11 | 2020-11-10 | 深圳墨世科技有限公司 | Article recommendation method, device and equipment based on article and storage medium |
CN112667845A (en) * | 2020-12-23 | 2021-04-16 | 腾讯音乐娱乐科技(深圳)有限公司 | Song recommendation method and device, electronic equipment and readable storage medium |
CN113672758A (en) * | 2021-08-20 | 2021-11-19 | 杭州网易云音乐科技有限公司 | Singing list generation method, device, medium and computing equipment |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113127676B (en) * | 2021-05-17 | 2022-07-01 | 杭州网易云音乐科技有限公司 | Information matching method, system, device, storage medium and electronic equipment |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101441667A (en) * | 2008-12-29 | 2009-05-27 | 北京搜狗科技发展有限公司 | Music recommend method and apparatus |
CN101626388A (en) * | 2009-07-24 | 2010-01-13 | 南京邮电大学 | Constructing method of incentive mechanism based on recommended node credibility computation |
CN102024058A (en) * | 2010-12-31 | 2011-04-20 | 万音达有限公司 | Music recommendation method and system |
CN102611785A (en) * | 2011-01-20 | 2012-07-25 | 北京邮电大学 | Personalized active news recommending service system and method for mobile phone user |
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 |
CN102929928A (en) * | 2012-09-21 | 2013-02-13 | 北京格致璞科技有限公司 | Multidimensional-similarity-based personalized news recommendation method |
CN103177722A (en) * | 2013-03-08 | 2013-06-26 | 北京理工大学 | Tone-similarity-based song retrieval method |
CN104991900A (en) * | 2015-06-09 | 2015-10-21 | 腾讯科技(深圳)有限公司 | Method and apparatus for pushing music data |
CN106095887A (en) * | 2016-06-07 | 2016-11-09 | 兰州大学 | Context aware Web service recommendation method based on weighted space-time effect |
-
2016
- 2016-12-28 CN CN201611239007.4A patent/CN108255840B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101441667A (en) * | 2008-12-29 | 2009-05-27 | 北京搜狗科技发展有限公司 | Music recommend method and apparatus |
CN101626388A (en) * | 2009-07-24 | 2010-01-13 | 南京邮电大学 | Constructing method of incentive mechanism based on recommended node credibility computation |
CN102024058A (en) * | 2010-12-31 | 2011-04-20 | 万音达有限公司 | Music recommendation method and system |
CN102611785A (en) * | 2011-01-20 | 2012-07-25 | 北京邮电大学 | Personalized active news recommending service system and method for mobile phone user |
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 |
CN102929928A (en) * | 2012-09-21 | 2013-02-13 | 北京格致璞科技有限公司 | Multidimensional-similarity-based personalized news recommendation method |
CN103177722A (en) * | 2013-03-08 | 2013-06-26 | 北京理工大学 | Tone-similarity-based song retrieval method |
CN104991900A (en) * | 2015-06-09 | 2015-10-21 | 腾讯科技(深圳)有限公司 | Method and apparatus for pushing music data |
CN106095887A (en) * | 2016-06-07 | 2016-11-09 | 兰州大学 | Context aware Web service recommendation method based on weighted space-time effect |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920689A (en) * | 2018-07-12 | 2018-11-30 | 江苏满运软件科技有限公司 | Source of goods recommended method and system |
CN109582817A (en) * | 2018-10-30 | 2019-04-05 | 努比亚技术有限公司 | A kind of song recommendations method, terminal and computer readable storage medium |
CN109543132A (en) * | 2018-11-22 | 2019-03-29 | 深圳墨世科技有限公司 | Content recommendation method, device, electronic equipment and storage medium |
CN110008369A (en) * | 2018-12-26 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Information processing method and its device, electronic equipment, computer-readable medium |
CN110457492A (en) * | 2019-06-06 | 2019-11-15 | 北京酷我科技有限公司 | A kind of individuation data proposed algorithm and recommender system |
CN110909199A (en) * | 2019-11-18 | 2020-03-24 | 彩讯科技股份有限公司 | Song recommendation display method and device based on vehicle-mounted system |
CN111915409A (en) * | 2020-08-11 | 2020-11-10 | 深圳墨世科技有限公司 | Article recommendation method, device and equipment based on article and storage medium |
CN111915409B (en) * | 2020-08-11 | 2023-08-04 | 深圳墨世科技有限公司 | Item recommending method, device, equipment and storage medium based on item |
CN112667845A (en) * | 2020-12-23 | 2021-04-16 | 腾讯音乐娱乐科技(深圳)有限公司 | Song recommendation method and device, electronic equipment and readable storage medium |
CN113672758A (en) * | 2021-08-20 | 2021-11-19 | 杭州网易云音乐科技有限公司 | Singing list generation method, device, medium and computing equipment |
CN113672758B (en) * | 2021-08-20 | 2024-04-12 | 杭州网易云音乐科技有限公司 | Song list generation method, device, medium and computing equipment |
Also Published As
Publication number | Publication date |
---|---|
CN108255840B (en) | 2021-04-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108255840A (en) | A kind of recommendation method and system of song | |
US20200372018A1 (en) | Query Results Ranked for Original & Real-Time Parameters | |
CN104899302B (en) | Recommend the method and apparatus of music to user | |
US8719283B2 (en) | Summarizing reviews | |
CN107885745A (en) | A kind of song recommendations method and device | |
CN108628833B (en) | Method and device for determining summary of original content and method and device for recommending original content | |
US11669296B2 (en) | Computerized systems and methods for hosting and dynamically generating and providing customized media and media experiences | |
CN105975472A (en) | Method and device for recommendation | |
CN109478195A (en) | The method and system of selection and optimization for search engine | |
CN107124653A (en) | The construction method of TV user portrait | |
CN107977373B (en) | Song recommendation method | |
Wall | Finding an alternative: Music programming in US college radio | |
WO2018121368A1 (en) | Method for generating music to accompany lyrics and related apparatus | |
CN108292314A (en) | Information processing unit, information processing method and program | |
CN107967280A (en) | A kind of method and system of label recommendations song | |
CN108471544A (en) | A kind of structure video user portrait method and device | |
CN108920585A (en) | The method and device of music recommendation, computer readable storage medium | |
CN105279289B (en) | Individualized music based on exponential damping window recommends sort method | |
Baur et al. | Listening factors: A large-scale principal components analysis of long-term music listening histories | |
CN105005576B (en) | A kind of video website similar users search system and method | |
CN103247318B (en) | A kind of obtain the method for video, device and play the method for video, device | |
CN109992694A (en) | A kind of music intelligent recommendation method and system | |
CN106294563B (en) | A kind for the treatment of method and apparatus of multi-medium data | |
CN107423307A (en) | The distribution method and device of a kind of internet information resource | |
Jun et al. | Social mix: automatic music recommendation and mixing scheme based on social network analysis |
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 |