CN107977370B - Singer recommendation method and system - Google Patents
Singer recommendation method and system Download PDFInfo
- Publication number
- CN107977370B CN107977370B CN201610920992.9A CN201610920992A CN107977370B CN 107977370 B CN107977370 B CN 107977370B CN 201610920992 A CN201610920992 A CN 201610920992A CN 107977370 B CN107977370 B CN 107977370B
- Authority
- CN
- China
- Prior art keywords
- singer
- song
- songs
- preference
- singers
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
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)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Reverberation, Karaoke And Other Acoustics (AREA)
Abstract
The invention relates to a singer recommendation method and system. The method comprises the following steps: and obtaining the preference score of the user playing the song according to the song listening behavior and the song listening source of the user, and further forming a song listening preference matrix of the user according to the preference score of the user playing the song. Wherein, the act of listening to songs comprises an act of actively listening to songs. Generating a playlist according to the active song listening behavior and the preference score of the user for playing songs, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity of each song in the playlist. Obtaining preference scores of singers according to preference scores of songs played by users, and further forming a singer preference matrix according to the preference scores of the singers; and selecting the singer according to the singer preference matrix, and calculating the similarity score of each singer in the selected singer to obtain the recommended singer.
Description
Technical Field
The invention relates to personalized recommendation of big data and machine learning algorithms, in particular to a singer recommendation method and system.
Background
In the prior art, user behavior data, song listening history and song characteristic data are very large and are finished by Hadoop.
Singer recommendations are mainly based on Item-Base collaborative filtering, which results in a large amount of noise data, resulting in very low singer similarity.
The recommended singer list has low relevance and irrelevant singers are ranked in the front, resulting in dissatisfaction of the user with the result.
Disclosure of Invention
The invention aims to solve the technical problems in the prior art and provides a singer recommendation method and system.
To achieve the above object, in one aspect, the present invention provides a singer recommendation method, including:
obtaining preference scores of songs played by a user according to the song listening behaviors and song listening sources of the user, and further forming a song listening preference matrix of the user according to the preference scores of the songs played by the user; wherein, the act of listening to songs comprises an act of actively listening to songs; generating a playlist according to the active song listening behavior and preference scores of songs played by a user, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity of each song in the playlist; obtaining preference scores of singers according to preference scores of songs played by users, and further forming a singer preference matrix according to the preference scores of the singers; and selecting the singer according to the singer preference matrix, and calculating the similarity score of each singer in the selected singer according to the singer similarity matrix to obtain the recommended singer.
Preferably, the step of calculating the preference score of the user for listening to the songs and recording the songs according to the user's song listening behavior and song listening source, and further forming the user song listening preference matrix according to the preference score further comprises: and updating the preference score of the song listened by the user according to the song time attenuation and the song popularity reduction, and further updating the song listening preference matrix of the user.
Preferably, the similarity between the singer corresponding to each song in the playlist and the singers corresponding to other songs in the playlist is calculated as:
wherein the content of the first and second substances,
a vector corresponding to the singer for each song in the playlist,
vectors corresponding to singers for other songs in the playlist.
Preferably, the method further comprises: and sorting the recommended singers according to a relevance sorting algorithm, and recommending the singers to the user.
Preferably, the method further comprises: and sorting the recommended singers according to the fitting degree and the threshold value of the prediction probability of the recommended singers and the target probability of the recommended singers, and recommending the singers to the user.
Preferably, the recommended predicted probability of the singer is:
wherein, PijPredicted probability of singer as recommended, siIs the similarity score, s, of the previously ranked singerjIs the similarity score of a singer who ranks behind, i is the singer who ranks ahead, and j is the singer who ranks behind.
Preferably, the recommended target probability for the singer is:
wherein the content of the first and second substances,
is the target probability, s, of the recommended singerij∈[-1,1]。
Preferably, the act of actively listening to songs comprises: one or more of focusing on songs, downloading songs, collecting songs, searching for songs, purchasing songs, and locally uploading songs.
In another aspect, the present invention provides a singer recommendation system, comprising: the device comprises a first unit, a second unit, a third unit and a selection unit. The first unit is used for obtaining preference scores of songs played by a user according to the song listening behaviors and song listening sources of the user and further forming a song listening preference matrix of the user according to the preference scores of the songs played by the user; wherein, the act of listening to songs comprises an act of actively listening to songs; the second unit is used for generating a playlist according to the active song listening behavior and the preference score of the user for playing songs, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity of each song in the playlist; the third unit is used for obtaining preference scores of the singers according to the preference scores of the songs played by the users, and further forming a singer preference matrix according to the preference scores of the singers; the selection unit is used for selecting the singer according to the singer preference matrix, and calculating and selecting the similarity score of each singer in the singer according to the singer similarity matrix to obtain the recommended singer.
Preferably, the system further comprises a refresh unit: the refreshing unit is used for updating preference scores of songs recorded by a user according to song time attenuation and song popularity reduction, and further updating a user song listening preference matrix.
Preferably, the system further comprises a fourth unit: and the system is used for sequencing the recommended singers according to the fitting degree and the threshold value of the prediction probability of the recommended singers and the target probability of the recommended singers, and recommending the singers to the user.
According to the singer recommendation method and system provided by the invention, singer similarity in a recommendation result is obviously improved; meanwhile, a relevance ranking algorithm based on machine learning is provided, so that singers with high relevance are ranked in front as much as possible and meet the requirements of users.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a flow chart illustrating a singer recommending method according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a singer recommendation system according to an embodiment of the present invention.
Detailed Description
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Fig. 1 is a schematic structural diagram of a singer recommendation method according to an embodiment of the present invention. As shown in fig. 1, the singer recommending method comprises the steps of:
step S100: obtaining preference scores of songs played by a user according to the song listening behaviors and song listening sources of the user, and further forming a song listening preference matrix of the user according to the preference scores of the songs played by the user; wherein, the act of listening to songs comprises an act of actively listening to songs;
and calculating the preference score of the song according to the song listening behavior and the song listening source of the user. The song listening behavior is divided into an active song listening behavior and a passive song listening behavior. The active action of listening to songs includes: heartburn, heartburn cancellation, downloading, collection, search, purchase, and local upload; passive act of listening to songs: playing and completely listening. The song listening sources comprise daily recommendations, song lists, radio stations, partitions and ranking lists.
For example, assuming that there are M users listening to songs, each user keeps the song of top3000 in the song listening record, so that a [ Mx 3000] dimension song listening preference matrix is obtained.
Step S110: generating a playlist according to the active song listening behavior and preference scores of songs played by a user, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity of each song in the playlist;
and generating a playlist from songs corresponding to the active song listening behavior in the user preference matrix according to the preference scores of the songs played by the user according to the active song listening behavior, such as the user hearts listing, searching, downloading and local uploading, and performing training by using a word2vec deep learning tool to obtain singer word vectors of singers corresponding to each song in the playlist.
The word2vec deep learning tool can simplify the processing of song text contents into vector operation in a K-dimensional vector space through training, and can further represent the semantic similarity of song texts, namely the similarity of singers, through the similarity on the vector space.
And calculating the similarity between the singer corresponding to one song in the playlist and the singers corresponding to other songs in the playlist by utilizing the cosine distance so as to obtain the similarity between the singer corresponding to one song in the playlist and the singers corresponding to other songs in the playlist, and combining the similarities of all the singers to form a singer similarity matrix.
For example, the top1000 singers closest to each singer are calculated as the similar singers of the singer. Assuming a total of N singers, after calculating the Similar singers of each singer, a singer similarity Matrix Artist Similar Matrix with [ Nx 1000] dimension is obtained.
And obtaining the similarity between one song in the playlist and other songs in the playlist through cosine similarity.
For example, assume that the lyrics track vector of a song in the playlist is
Other lyrics vectors of the playlist are
Theta is
And
the previous angle, then the cosine theorem rewrites the following form:
step S120: obtaining preference scores of singers according to preference scores of songs played by users, and further forming a singer preference matrix according to the preference scores of the singers;
step S130: and selecting the singer according to the singer preference matrix, and calculating the similarity score of each singer in the selected singer according to the singer similarity matrix to obtain the recommended singer.
For example, the preset rule selects 50 favorite singers from the singer preference matrix similarity matrix.
50 favorite 50 singers [ artist01, artist02, …, artist50] song01, song02, …, song50] were selected from the playlist. For a top50 song singer, the similarity of the song singer's song and other song singers in the playlist is selected from the song singer preference matrix similarity matrix. [ artist01_ similar 01, artist01_ similar 02, artist01_ similar 03, … ] [ song01_ similar 01, song01_ similar 02, song01_ similar 03, … ] is the similarity of the first artist in top50 to the other artists.
[ artist02_ similar 01, artist02_ similar 02, artist02_ similar 03, … ] is the similarity of the second artist to the other artists in top 50.
[ artist50_ similar 01, artist50_ similar 02, artist50_ similar 03, … ] is the similarity of the 50 th singer in top50 to the other singers.
And accumulating the scores of the similar singers to be finally recommended to the user.
Optionally, the preference score calculated in step S100 is not constant, and decreases with time and whether it is the hottest song, that is, the song preference matrix also changes.
Step S140 is also included after step S100:
step S140: and updating the preference score of the song listened by the user according to the song time attenuation and the song popularity reduction, and further updating the song listening preference matrix of the user.
Specifically, the preference score of the song needs to be updated by considering two factors of time attenuation and heat weight reduction. The rule for the drop weight is that as time goes on, the song preference score decreases; as the popularity increases, the song preference score increases.
Optionally, the singer recommending method further comprises:
step S150: and the system is used for sequencing the recommended singers according to the fitting degree and the threshold value of the prediction probability of the recommended singers and the target probability of the recommended singers, and recommending the singers to the user.
In step S130, a set of singers and singer similarity scores of top50 have been obtained.
For example, [ artist01:10, artist02:8, artist03:5, …, artist50:12 ].
RankNet is an implementation of the LambdaMART model, a pairwise model, which transforms the ranking problem into a ranking probability problem that compares an (i, j) pair, i.e., compares the probability that artist _ i is ranked ahead of artist _ j. It is first rootedCalculating the predicted probability P of the singer's pair according to the similarity score of each singerij：
Wherein s isiMeans similarity score, s, of artist _ ijRefers to the similarity score of artist _ j.
The RankNet is used for sequencing a singer sequence, only the sequencing probability between adjacent singers needs to be calculated, all pair does not need to be calculated, and the calculation amount is reduced.
Through this step, a probability sequence [ p12:0.3, p23:0.1, p34:0.5, …, pk-1k:0.6] is obtained
Pij>0.5 indicates that i should be ranked in front of j.
In the last step, a new singer sequence is obtained according to the probability value and is recorded as a predicted sequencing sequence, and the target probability value is calculated according to the sequence:
wherein SijTaking {1,0, -1}, and respectively corresponding to three conditions that the artist _ i is higher than the artist _ j in sequence, the artist _ i is the same as the artist _ j in sequence, and the artist _ i is lower than the artist _ j. Thus, based on the labeling scores of artist _ i and artist _ j, the target probability of { artist _ i, artist _ j } can be calculated.
Through this step, the probability is obtained
Sequence of [ p12:1, p23:1, p34:0.5, …, pk-1k:0]
P is then measured using cross entropy as a loss functionijTo pair
I.e. the cost c (cost) of calculating the predicted ordered sequence.
According to corresponding PijAnd
can calculate a pair of singer sequences<artist_i,artist_j>When the cost value of cost is larger than a threshold, the sorting cost is too high, and the original sequence should be maintained; cost<threshold, which means that the cost is not so high, should be ordered in the order in the prediction sequence. threshold is obtained from empirical values.
Thus, the order of arrangement between the final singers is determined according to the cost.
Fig. 2 is a schematic structural diagram of a station recommendation system according to an embodiment of the present invention. As shown in fig. 2, the station recommendation system includes: a first unit 10, a second unit 20, a third unit 30 and a selection unit 40.
The first unit 10 is configured to obtain a preference score of a user for playing a song according to a song listening behavior and a song listening source of the user, and further form a song listening preference matrix of the user according to the preference score of the user for playing the song; wherein, the act of listening to songs comprises an act of actively listening to songs; the second unit 20 is configured to generate a playlist according to the active song listening behavior and the preference score of the user for playing songs, obtain a singer word vector of a singer corresponding to each song in the playlist by using a deep learning tool, calculate a similarity between the singer corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and form a singer similarity matrix according to the similarity between each song in the playlist; the third unit 30 is configured to obtain a preference score of the singer according to the preference score of the song played by the user, and further form a singer preference matrix according to the preference score of the singer; the selecting unit 40 is configured to select a singer according to the singer preference matrix, and calculate a similarity score for each singer in the selected singer according to the singer similarity matrix, so as to obtain a recommended singer.
Optionally, the calculated preference score is not constant and may vary with time and popularity of the song, as may the corresponding song preference matrix.
Specifically, the singer recommendation system further includes a refresh unit 50:
the refreshing unit 50 is used for updating the preference score of the song recorded by the user according to the song time attenuation and the song popularity reduction, and further updating the user song listening preference matrix.
Optionally, the comprehensive information of the user can better reflect the preference of the user, and further can select songs matched with the user.
Specifically, the singer recommendation system further includes a fourth unit 60:
the fourth unit 60 is configured to rank the recommended singers and recommend them to the user based on a fitting degree and a threshold value comparing the predicted probability of the recommended singers and the target probability of the recommended singers.
The singer recommendation method and the system provided by the embodiment of the invention obviously improve the singer similarity in the recommendation result; meanwhile, a relevance ranking algorithm based on machine learning is provided, so that singers with high relevance are ranked in front as much as possible and meet the requirements of users.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A singer recommendation method, comprising:
obtaining preference scores of songs played by a user according to the song listening behaviors and song listening sources of the user, and further forming a song listening preference matrix of the user according to the preference scores of the songs played by the user; wherein the act of listening to songs comprises an act of actively listening to songs; the song listening sources comprise daily recommendation, a song list, a radio station, a partition and a ranking list;
generating a playlist according to the active song listening behavior and the preference score of the user for playing songs, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity of each song in the playlist;
obtaining preference scores of singers according to the preference scores of songs played by the user, and further forming a singer preference matrix according to the preference scores of the singers;
and selecting singers according to the singer preference matrix, and calculating the similarity score of each singer in the selected singers according to the singer similarity matrix to obtain recommended singers.
2. The method according to claim 1, wherein the step of calculating the preference score of the user for listening to the recorded songs according to the user's behavior and source of listening to the songs and further forming the preference matrix according to the preference score further comprises:
and updating the preference score of the song listened by the user according to the song time attenuation and the song popularity reduction, and further updating the song listening preference matrix of the user.
3. The method of claim 2, wherein the calculating the similarity between the singer corresponding to each song in the playlist and the singers corresponding to other songs in the playlist is:
wherein the content of the first and second substances,
a vector corresponding to the singer for each song in the playlist,
for the others in the playlistThe song corresponds to the vector of the singer.
4. The method of claim 3, further comprising:
and sequencing the recommended singers according to the fitting degree and the threshold value of the prediction probability of the recommended singers and the target probability of the recommended singers, and recommending the singers to the user.
5. The method of claim 4, wherein the predicted probability of the recommended singer is:
wherein, PijFor the predicted probability, s, of the recommended singeriIs the similarity score, s, of the previously ranked singerjIs the similarity score of a singer who ranks behind, i is the singer who ranks ahead, and j is the singer who ranks behind.
6. The method of claim 4, wherein the target probability of the recommended singer is:
wherein the content of the first and second substances,
for the target probability, S, of the recommended singerijTake {1,0, -1 }.
7. The method of claim 1, wherein the act of actively listening to songs comprises: one or more of focusing on songs, downloading songs, collecting songs, searching for songs, purchasing songs, and locally uploading songs.
8. A singer recommendation system, comprising: a first unit (10), a second unit (20), a third unit (30) and a selection unit (40); wherein the content of the first and second substances,
the device comprises a first unit (10) and a second unit, wherein the first unit is used for obtaining preference scores of songs played by a user according to the song listening behaviors and song listening sources of the user and further forming a song listening preference matrix of the user according to the preference scores of the songs played by the user; wherein the act of listening to songs comprises an act of actively listening to songs; the song listening sources comprise daily recommendation, a song list, a radio station, a partition and a ranking list;
a second unit (20) for generating a playlist according to the active song listening behavior and the preference score of the user for playing songs, obtaining singer word vectors of singers corresponding to each song in the playlist by using a deep learning tool, calculating the similarity between the singers corresponding to each song in the playlist and the singers corresponding to other songs in the playlist, and forming a singer similarity matrix according to the similarity between each song in the playlist;
a third unit (30) for obtaining a preference score of the singer according to the preference score of the song played by the user, and further forming a singer preference matrix according to the preference score of the singer;
and the selecting unit (40) is used for selecting the singers according to the singer preference matrix and calculating the similarity score of each singer in the selected singers according to the singer similarity matrix to obtain the recommended singer.
9. The system according to claim 8, characterized in that it further comprises a refresh unit (50):
and the refreshing unit (50) is used for updating the preference score of the song recorded by the user according to the song time attenuation and the song popularity reduction, and further updating the user song listening preference matrix.
10. The system according to claim 9, characterized in that it further comprises a fourth unit (60):
the fourth unit (60) is used for sorting the recommended singers according to the fitting degree and the threshold value of the prediction probability of the recommended singer and the target probability of the recommended singer, and recommending the singers to the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610920992.9A CN107977370B (en) | 2016-10-21 | 2016-10-21 | Singer recommendation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610920992.9A CN107977370B (en) | 2016-10-21 | 2016-10-21 | Singer recommendation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107977370A CN107977370A (en) | 2018-05-01 |
| CN107977370B true CN107977370B (en) | 2021-03-12 |
Family
ID=62003680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610920992.9A Active CN107977370B (en) | 2016-10-21 | 2016-10-21 | Singer recommendation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107977370B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635153B (en) * | 2018-11-22 | 2023-01-13 | 腾讯音乐娱乐科技（深圳）有限公司 | Migration path generation method, device and storage medium |
| CN110083772A (en) * | 2019-04-29 | 2019-08-02 | 北京小唱科技有限公司 | Singer's recommended method and device based on singing skills |
| CN110222233B (en) * | 2019-06-14 | 2021-01-15 | 北京达佳互联信息技术有限公司 | Video recommendation method and device, server and storage medium |
| CN111611432B (en) * | 2020-05-29 | 2023-09-15 | 北京酷我科技有限公司 | Singer classification method based on Labeled LDA model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103119905A (en) * | 2010-09-09 | 2013-05-22 | 阿尔卡特朗讯 | Method and apparatus for targeted communications |
| CN103885987A (en) * | 2012-12-21 | 2014-06-25 | 中国移动通信集团公司 | Music recommendation method and system |
| CN104054104A (en) * | 2011-11-16 | 2014-09-17 | 谷歌股份有限公司 | Displaying auto-generated facts about a music library |
| CN104573105A (en) * | 2015-01-30 | 2015-04-29 | 福州大学 | Method of recommending hit songs and singers in music on-demand network |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009210790A (en) * | 2008-03-04 | 2009-09-17 | Nec Software Kyushu Ltd | Music selection singer analysis and recommendation device, its method, and program |
| CN101984437B (en) * | 2010-11-23 | 2012-08-08 | 亿览在线网络技术（北京）有限公司 | Music resource individual recommendation method and system thereof |
| CN104899302B (en) * | 2015-06-10 | 2018-07-17 | 百度在线网络技术（北京）有限公司 | Recommend the method and apparatus of music to user |
| CN105575393A (en) * | 2015-12-02 | 2016-05-11 | 中国传媒大学 | Personalized song recommendation method based on voice timbre |
-
2016
- 2016-10-21 CN CN201610920992.9A patent/CN107977370B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103119905A (en) * | 2010-09-09 | 2013-05-22 | 阿尔卡特朗讯 | Method and apparatus for targeted communications |
| CN104054104A (en) * | 2011-11-16 | 2014-09-17 | 谷歌股份有限公司 | Displaying auto-generated facts about a music library |
| CN103885987A (en) * | 2012-12-21 | 2014-06-25 | 中国移动通信集团公司 | Music recommendation method and system |
| CN104573105A (en) * | 2015-01-30 | 2015-04-29 | 福州大学 | Method of recommending hit songs and singers in music on-demand network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107977370A (en) | 2018-05-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kula | Metadata embeddings for user and item cold-start recommendations | |
| Weston et al. | Label partitioning for sublinear ranking | |
| US10516906B2 (en) | Systems, methods, and computer products for recommending media suitable for a designated style of use | |
| US10423648B2 (en) | Method, system, and computer readable medium for interest tag recommendation | |
| US8554854B2 (en) | Systems and methods for identifying terms relevant to web pages using social network messages | |
| CN108009181B (en) | Radio station recommendation method and system | |
| CN107977370B (en) | Singer recommendation method and system | |
| CN106777051A (en) | A kind of many feedback collaborative filtering recommending methods based on user's group | |
| CN106202294B (en) | Related news computing method and device based on keyword and topic model fusion | |
| CN110175895B (en) | Article recommendation method and device | |
| WO2011034475A1 (en) | Method and apparatus for executing a recommendation | |
| EP2783303A1 (en) | Prototype-based re-ranking of search results | |
| JP6137960B2 (en) | Content search apparatus, method, and program | |
| Ma et al. | Your Tweets Reveal What You Like: Introducing Cross-media Content Information into Multi-domain Recommendation. | |
| US20130332462A1 (en) | Generating content recommendations | |
| KR20130091392A (en) | Apparatus and method for recommending keyword | |
| Rubtsov et al. | A hybrid two-stage recommender system for automatic playlist continuation | |
| CN109299357B (en) | Laos language text subject classification method | |
| Shah et al. | Prompt: Personalized user tag recommendation for social media photos leveraging personal and social contexts | |
| Hawashin et al. | An efficient user interest extractor for recommender systems | |
| EP3144825A1 (en) | Enhanced digital media indexing and retrieval | |
| US20170075999A1 (en) | Enhanced digital media indexing and retrieval | |
| Sánchez-Moreno et al. | Dynamic inference of user context through social tag embedding for music recommendation | |
| Diaz-Aviles et al. | Swarming to rank for recommender systems | |
| CN108932262B (en) | Song recommendation method and device |
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 |