CN103324686B - Real time individual video recommendation method based on text flow network - Google Patents
Real time individual video recommendation method based on text flow network Download PDFInfo
- Publication number
- CN103324686B CN103324686B CN201310217181.9A CN201310217181A CN103324686B CN 103324686 B CN103324686 B CN 103324686B CN 201310217181 A CN201310217181 A CN 201310217181A CN 103324686 B CN103324686 B CN 103324686B
- Authority
- CN
- China
- Prior art keywords
- user
- vector
- long
- term interest
- video
- 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
- 238000000034 method Methods 0.000 title claims abstract description 22
- 230000007774 longterm Effects 0.000 claims abstract description 63
- 239000000203 mixture Substances 0.000 claims abstract description 32
- 230000002452 interceptive effect Effects 0.000 claims description 4
- 230000006399 behavior Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 206010068052 Mosaicism Diseases 0.000 description 1
- 230000018199 S phase Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000003765 sex chromosome Anatomy 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The step of present invention real time individual based on text flow network video recommendation method is as follows: active user document set up in S1 user's current literary composition that pushes away issued and forward on text flow network, multiple active user documents are set up focus incident space, obtains multiple focus incidents that user is currently paid close attention to；The S2 user's all information on text flow network set up customer documentation, multiple customer documentations are set up Long-term Interest theme space, obtains multiple user respectively in the Long-term Interest distribution vector in this space；The plurality of focus incident is ranked up by S3 user's Long-term Interest distribution vector, it is thus achieved that user is currently most interested in focus incident；S4 retrieves at Video Applications platform and is currently most interested in multiple videos that focus incident is relevant to user；S5 at Video Applications platform information, obtains user's Long-term Interest characteristic vector with user；The plurality of video is reordered by S6 user's Long-term Interest characteristic vector, and gives this user top n video recommendations.
Description
Technical field
The present invention relates to interconnect personalized recommendation technical field, particularly relate to a kind of real-time individual character based on text flow network
Change video recommendation method.
Background technology
Along with the development of Internet technology, the arrival in particularly WEB2.0 epoch, the propagation of Online Video has reached
Unprecedented level.The video data of magnanimity can meet the demand of nearly all user even so, but also make to search simultaneously
Seek and find user's video interested and become a thing the most loaded down with trivial details.Therefore, individualized video recommendation is right
Being very important now in information overload.
Traditional individualized video recommends method to be based on static subscriber's model, and this model utilizes user's registration information, goes through
History behavior understands the Long-term Interest of user.But, the renewal of current information is more and more frequent.User is facing a large amount of for every day
Fresh information, causes the short-term interest of user along with current hotspot event is in continuous drift.Such as, the U.S. is read as a user
During the news of presidential inauguration ceremony, he is likely to removal search associated video and goes to further appreciate that this event.Perhaps from the most emerging
From the perspective of interest, this user is not interested to politics, but his short-term interest is but by current hotspot event
Have impact on.In this case, traditional individualized video recommends method to tackle because they cannot to catch user emerging
The drift of interest.
Fig. 1 is the flow chart that prior art carries out individualized video recommendation.As it is shown in figure 1, prior art carries out personalization
The flow process of video recommendations includes:
Step S101, utilizes user to set up user in the information (such as log-on message, historical behavior) of some network platform
Long-term Interest model, usually characteristic vector, a certain preference of every one-dimensional representation user；
Step S102, utilizing video information (such as video labeling, contextual information and video content) is that each video is built
Vertical characteristic vector；
Step S103, utilizes user's special case vector to be ranked up video with the inner product of video feature vector, and score is high
The video recommendations of (such as front 10) is to user.
Inventor finds that the method that above-mentioned individualized video is recommended exists following technological deficiency:
1) Long-term Interest of user is focused on, it is impossible to the short-term interest preference of real-time capture user；
2) only utilize single network platform information to learn user interest, often there are cold start-up (cold-start) and data
Openness (data sparsity) problem.
Summary of the invention
(1) to solve the technical problem that
For solving above-mentioned problem, the invention provides a kind of real time individual video recommendations side based on text flow network
Method, to improve the accuracy that individualized video is recommended.
(2) technical scheme
The present invention provides real time individual video recommendation method based on text flow network, and described individualized video is recommended
Step includes:
Step S1: utilize user's literary composition that pushes away that current institute issues and forwards on text flow network to set up active user document,
Utilization pushes away special potential Di Likeli distributed model and multiple active user documents is set up focus incident space, and obtains multiple user
Respectively in the distribution vector in this focus incident space, i.e. obtain multiple focus incidents that user is currently paid close attention to；
Step S2: utilize all literary compositions that push away of user's log-on message on text flow network and issue and forwarding to set up user
Document, utilizes topic model that multiple customer documentations set up a Long-term Interest theme space, and obtains multiple user and exist respectively
This Long-term Interest theme space respective Long-term Interest distribution vector；
Step S3: utilize user in the long-term interest topic of text flow network Long-term Interest distribution vector spatially to user
The current multiple focus incidents paid close attention to are ranked up, it is thus achieved that the focus incident that user is currently most interested in；
Step S4: retrieve at Video Applications platform and be currently most interested in multiple videos that focus incident is relevant to user；
Step S5: utilize user the log-on message of Video Applications platform and with the interactive information of video, set up user and exist
The Long-term Interest vector space model of Video Applications platform, obtains user's Long-term Interest characteristic vector at Video Applications platform；
Step S6: utilize user in the Long-term Interest characteristic vector of Video Applications platform to the multiple videos described in step S4
Reorder, and give this user top n video recommendations.
(3) beneficial effect
From technique scheme it can be seen that present invention real time individual based on text flow network video recommendation method has
There is a following beneficial effect:
(1) rapidity that text flow network hotspot event occurs and propagates is utilized, the focus that detection user is paid close attention in real time
Event, effectively captures the short-term interest of user, thus improves the accuracy that individualized video is recommended；
(2) make use of user at the information learning user interest of different platform, effectively alleviate cold start-up and data are dilute
Dredge sex chromosome mosaicism.
Accompanying drawing explanation
Fig. 1 is that prior art utilizes traditional method to carry out the flow chart of individualized video recommendation；
Fig. 2 is the flow chart of embodiment of the present invention real time individual based on text flow network video recommendation method.
Detailed description of the invention
It should be noted that in accompanying drawing or description describe, similar or identical part all uses identical figure number.And
In the accompanying drawings, to simplify or convenient sign.Furthermore, the implementation not illustrating in accompanying drawing or describing, for art
In form known to a person of ordinary skill in the art.It addition, although the demonstration of the parameter comprising particular value can be provided herein, it is to be understood that
Parameter is worth equal to corresponding without definite, but can be similar to be worth accordingly in acceptable error margin or design constraint.
It is an object of the invention to realize real time individual video recommendations.There is following challenge in this problem.First, we are difficult to
Accurately capture the short-term interest of user；It addition, user is the most limited at the available information of single platform, it is difficult to accurate assurance user
Long-term Interest；Finally, the short-term interest and the Long-term Interest that how to merge user are also difficult points.
It should be noted that it will be understood by those skilled in the art that above-mentioned video can also be audio frequency, picture etc.,
Hereinafter mainly illustrate as a example by video, but the invention is not limited in this.It addition, above-mentioned text flow network is hereinafter
Illustrating as a example by pushing away spy (Twitter), Video Applications platform illustrates as a example by excellent prominent rich (YouTube), but this
Bright it is not limited thereto.
In one exemplary embodiment of the present invention, it is proposed that a kind of real time individual video based on text flow network
The method recommended.As in figure 2 it is shown, the present embodiment real time individual based on text flow network video recommendation method includes:
Step S1: utilize user's literary composition that pushes away that current institute issues or forwards on text flow network to set up active user document,
Utilization pushes away special potential Di Likeli distributed model and multiple active user documents is set up focus incident space, and obtains multiple user
Respectively in the distribution vector in this focus incident space, i.e. obtain multiple focus incidents that user is currently paid close attention to；
User issues or forwards and push away literary composition (tweet) pushing away special platform, and these behaviors are by current hotspot event, Yong Huhao
Friend and the coefficient result of the interest of user own, be a kind of embodiment of user's short-term interest.Owing to pushing away the short text characteristic of literary composition,
Every pushes away literary composition and mainly expresses an event, and therefore, we use and push away special potential Di Likeli distributed model (TwitterLDA)
Focus incident space set up in the multiple literary compositions that push away currently issued from multiple users or forward, and each focus incident is by some semantic words
The vector that remittance is constituted, the probability that a certain semantic vocabulary of every one-dimensional representation of vector occurs in this event.But due to push away literary composition and
Containing much noise in log-on message, such as insignificant vocabulary and erroneous input.Therefore we use word net to filter.
Based on foregoing description, described in step S1, utilize what user current institute on text flow network issued or forwarded to push away literary composition
Set up specifically comprising the following steps that of active user document
Step S1a: collect from network multiple user issue the most respectively and forward push away literary composition；
Step S1b: utilize the above-mentioned noise pushed away in literary composition of word net filtration, pushes away literary composition after being filtered；
Step S1c: for each in multiple users, active user literary composition set up respectively in the literary composition that pushes away after utilizing it to filter
Shelves.
The distribution vector of each user has described in step S1 nonzero element, the focus thing that described nonzero element is corresponding
Part is the focus incident that user is currently paid close attention to, and therefore we have obtained multiple focus incidents that user is currently paid close attention to.
Above-mentioned " currently " can be " in one hour ", and " in one day " arbitrarily can embody the time range of real-time.
Step S2: utilize all literary compositions that push away of user's log-on message on text flow network and issue and forwarding to set up user
Document, utilizes topic model that multiple customer documentations set up a Long-term Interest theme space, and obtains multiple user and exist respectively
This Long-term Interest theme space respective Long-term Interest distribution vector；
User has reacted the Long-term Interest of user at the log-on message pushed away on spy；Meanwhile, the institute that user issued and forwarded
Have and push away literary composition and can react the Long-term Interest of user.Therefore we utilize the log-on message of user and user to issue and forward push away literary composition
Set up customer documentation.But the document contains much noise, such as insignificant vocabulary and erroneous input.Here we use equally
Word net filters.
Based on foregoing description, step S2 utilizes user's information on text flow network to set up the step of customer documentation such as
Under:
Step S2a: collect from network multiple user issue respectively and forward push away literary composition and log-on message thereof；
Step S2b: utilize the above-mentioned noise pushed away in literary composition and log-on message of word net filtration, filter except pushing away in literary composition and log-on message
Noun composition outside composition, push away literary composition and log-on message after being filtered；
Step S2c: for each user in multiple users, utilize each user to issue and forward push away literary composition and note
Noun composition in volume information sets up each customer documentation respectively.
Step S2: described in topic model can select potential Di Likeli distributed model (LDA), naturally it is also possible to select
Other models well known in the art, such as: the potential doctrine of probability analyzes model (PLSA) or turbine topic model (Turbo
Topic)。
Step S3: utilize user in the long-term interest topic of text flow network Long-term Interest distribution vector spatially to user
The current multiple focus incidents paid close attention to are ranked up, it is thus achieved that the focus incident that user is currently most interested in；
User currently may pay close attention to multiple focus incident, and we speculate use by the Long-term Interest distribution vector of user
The focus incident that family is most interested.Being ranked up the plurality of focus incident, first, we pass through relative entropy (Relative
Entropy) each focus incident of currently being paid close attention to of user and each theme similar in user's Long-term Interest theme space are calculated
Degree, then calculates user in conjunction with user's Long-term Interest distribution vector in Long-term Interest theme space every currently paid close attention to
Interest score value in individual event, the event of score value maximum is the focus incident that user is currently most interested in；Described user is current
The each focus incident paid close attention to and the similarity of each theme, described average relative entropy in user's Long-term Interest theme space
D (z | | x) it is expressed as follows:
Theme vector during wherein z is Long-term Interest theme space, x is current hotspot event space focus incident vector, D
(z | | x) represent the average relative entropy between theme vector z and focus incident vector x, z (i) and x (i) represent theme vector z and
Focus incident vector x probit on i-th semantic vocabulary, K is the dimension of lexical space, i=1,2 ... K.Average relative
The inverse of entropy is theme vector z and the similarity of focus incident vector x.
Described user interest score value p in each event currently paid close attention to (x | u, λ) it is expressed as follows:
The a certain focus incident vector that wherein x is currently paid close attention to by user u；λ be user u at Long-term Interest theme spatially
Distribution vector；Φ is Long-term Interest theme space；λZRepresent that user u is in Long-term Interest theme space Φ on theme vector z
Probit；P (x | u, λ) represent under given user u and distribution vector λ on Long-term Interest theme space Φ thereof, focus
The score of event vector x；Calculate the score of the multiple focus incidents currently paid close attention to respectively by user u after, we are again score
The focus incident that the highest focus incident is currently most interested in as user.
Through step S3, we achieve the rapidity utilizing text flow network hotspot event to occur and propagate, examine in real time
Survey the focus incident that user is currently most interested in, effectively capture the short-term interest of user.
Step S4: retrieve at Video Applications platform and be currently most interested in multiple videos that focus incident is relevant to user；
As described in step S2, the vector that focus incident is made up of some semantic vocabulary, each of vector represents a certain language
The probability that justice vocabulary occurs in this event.Therefore, during we select the focus incident that user is currently most interested in, probability of occurrence is
Three big semantic vocabulary, as query word, are retrieved associated video at Video Applications platform, and are selected front 20 to regard to first 100
Frequently, embodiment is chosen front 20 or front 50 or front 100 videos.
Step S5: utilize user the log-on message of Video Applications platform and with the interactive information of video, set up user and exist
The Long-term Interest vector space model of Video Applications platform, obtains user's Long-term Interest characteristic vector at Video Applications platform；
User has reacted the Long-term Interest of user, meanwhile, the user master to video at the log-on message of Video Applications platform
Dynamic behavior (such as upload or collect) reflects the Long-term Interest hobby of user.Therefore we utilize the log-on message of user, and
User uploads or collects the semantic label of video, classification and description and sets up user's Long-term Interest vector at Video Applications platform
Spatial model.
Utilize user the log-on message of Video Applications platform and concrete with the interactive information of video described in step S5
Step is as follows:
Step S5a: collect user's registration information and the semantic label of video, classification and the description uploading or collect；
Step S5b: utilize the noise in the above-mentioned log-on message of word net filtration and semantic label, classification and description, filter institute
State the composition removed name from the rolls in log-on message and semantic label, classification and description outside word composition；
Setting up user described in step S5 at the Long-term Interest vector space model of Video Applications platform is: utilize described note
Volume information and semantic label, classification and description in noun composition set up each user's Long-term Interest vector space model, obtain
User is in the Long-term Interest characteristic vector of Video Applications platform.
Step S6: utilize user in the Long-term Interest characteristic vector of Video Applications platform to the multiple videos described in step S4
Reorder, and give this user top n video recommendations.
First, we utilize the semantic label of video, classification and are described as each video and set up characteristic vector, then we
In the matching degree of the Long-term Interest characteristic vector of this Video Applications platform, video is arranged with user according to this feature vector
Sequence, specifically includes, and a given video υ, user is in Long-term Interest characteristic vector θ of Video Applications platform, the score of this video υ
It is expressed as:
Wherein M represents the dimension of vector space, i=1,2 ... M；P (υ | θ) represent given user at Video Applications platform
Long-term Interest characteristic vector θ under, the score of video υ；υiExpression is video feature vector weight in i-th dimension, θiRepresent and use
Head of a household's phase interest characteristics vector weight in i-th dimension.Calculate after score for each video, we further according to these scores pair with
Relevant front 20 the reordering to front 100 videos, then front 5 to first 20 of the focus incident that user is currently most interested in
Associated video recommends user.In embodiment and choose front 20 or front 50 or front 100 videos and reorder, then front
5 or front 10 or front 20 associated videos recommend user.
N in above-mentioned N number of video can be the most rationally to count, such as 5 to 20, simply need to be less than in step S4 selected many
The number of individual video.
Described Video Applications platform is the Video Applications platforms such as excellent cruel (YouKu), excellent prominent rich (YouTube), in embodiment
Illustrate as a example by excellent prominent rich (YouTube).Described associated video is front 10 or the most rationally number associated video recommendation
To this user.
Particular embodiments described above, has been carried out the purpose of the present invention, technical scheme and beneficial effect the most in detail
Describe in detail bright, be it should be understood that the specific embodiment that the foregoing is only the present invention, be not limited to the present invention, all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included in the guarantor of the present invention
Within the scope of protecting.
Claims (6)
1. the method for a real time individual video recommendations based on text flow network, it is characterised in that realize individualized video and push away
The step recommended includes:
Step S1: utilize user's literary composition that pushes away that current institute issues and forwards on text flow network to set up active user document, utilize
Push away special potential Di Likeli distributed model and multiple active user documents are set up focus incident space, and obtain multiple user respectively
In the distribution vector in this focus incident space, i.e. obtain multiple focus incidents that user is currently paid close attention to；
Step S2: utilize all literary compositions that push away of user's log-on message on text flow network and issue and forwarding to set up user's literary composition
Shelves, utilize topic model that multiple customer documentations are set up a Long-term Interest theme space, and obtain multiple user respectively at this
Long-term Interest theme space respective Long-term Interest distribution vector；
Step S3: utilize user current to user in the long-term interest topic of text flow network Long-term Interest distribution vector spatially
The multiple focus incidents paid close attention to are ranked up, it is thus achieved that the focus incident that user is currently most interested in；
Step S4: retrieve at Video Applications platform and be currently most interested in multiple videos that focus incident is relevant to user；
Step S5: utilize user the log-on message of Video Applications platform and with the interactive information of video, set up user at video
The Long-term Interest vector space model of application platform, obtains user's Long-term Interest characteristic vector at Video Applications platform；
Step S6: utilize user in the Long-term Interest characteristic vector of Video Applications platform, the multiple videos described in step S4 to be carried out
Reorder, and give this user top n video recommendations.
2. the method for real time individual video recommendations based on text flow network as claimed in claim 1, it is characterised in that each
Focus incident is the vector being made up of multiple semantic vocabulary, and a certain semantic vocabulary of every one-dimensional representation of vector occurs in this event
Probability.
3. the method for real time individual video recommendations based on text flow network as claimed in claim 1, it is characterised in that to institute
State multiple focus incident and be ranked up being to calculate each focus incident and the user's Long-term Interest theme sky that user is currently paid close attention to
The similarity of each theme between, then in conjunction with user's Long-term Interest distribution vector in Long-term Interest theme space, calculates and uses
Family interest score value in each event currently paid close attention to, the event of score value maximum is the focus that user is currently most interested in
Event.
4. the method for real time individual video recommendations based on text flow network as claimed in claim 3, it is characterised in that use
Average relative entropy calculates each focus incident and each theme in user's Long-term Interest theme space that described user is currently paid close attention to
Similarity, described average relative entropy D (z | | x) is expressed as:
Theme vector during wherein z is Long-term Interest theme space, x is current hotspot event space focus incident vector, D (z | |
X) represent that the average relative entropy between theme vector z and focus incident vector x, z (i) and x (i) represent theme vector z and focus
Event vector x probit on i-th semantic vocabulary, K is the dimension of lexical space, i=1,2 ... K；Average relative entropy
Inverse is theme vector z and the similarity of focus incident vector x.
5. the method for real time individual video recommendations based on text flow network as claimed in claim 3, it is characterised in that user
Interest score value p in each event currently paid close attention to (x | u, λ) it is expressed as follows:
Wherein: a certain focus incident vector that x is currently paid close attention to by user u, λ be user u at Long-term Interest theme spatially
Distribution vector, Φ is user's u Long-term Interest theme space；λZRepresent user u probit on theme vector z；P (x | u, λ)
Represent under given user u and distribution vector λ on Long-term Interest theme space Φ thereof, the score of focus incident vector x.
6. the method for real time individual video recommendations based on text flow network as claimed in claim 1, it is characterised in that described
Reordering multiple videos is to reorder to front 100 videos to front 20, then the most relevant regards to first 20 front 5
Frequency recommends user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310217181.9A CN103324686B (en) | 2013-06-03 | 2013-06-03 | Real time individual video recommendation method based on text flow network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310217181.9A CN103324686B (en) | 2013-06-03 | 2013-06-03 | Real time individual video recommendation method based on text flow network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103324686A CN103324686A (en) | 2013-09-25 |
| CN103324686B true CN103324686B (en) | 2016-12-28 |
Family
ID=49193429
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310217181.9A Active CN103324686B (en) | 2013-06-03 | 2013-06-03 | Real time individual video recommendation method based on text flow network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103324686B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915345B (en) * | 2014-03-11 | 2018-08-21 | 华为技术有限公司 | A kind of method and terminal of recommendation service information |
| CN105100165B (en) * | 2014-05-20 | 2017-11-14 | 深圳市腾讯计算机系统有限公司 | Network service recommends method and apparatus |
| CN104951563A (en) * | 2015-07-08 | 2015-09-30 | 北京理工大学 | Method and device for determining to-be-recommended objects |
| CN105045868B (en) * | 2015-07-14 | 2019-07-02 | 无锡天脉聚源传媒科技有限公司 | A kind of method and device for searching for hot ticket |
| CN107807964B (en) * | 2017-10-11 | 2021-02-09 | 咪咕互动娱乐有限公司 | Digital content ordering method, apparatus and computer readable storage medium |
| CN110418200A (en) * | 2018-04-27 | 2019-11-05 | Tcl集团股份有限公司 | A kind of video recommendation method, device and terminal device |
| CN111324723B (en) * | 2018-12-13 | 2023-04-25 | 阿里巴巴集团控股有限公司 | Sentence selection method, sentence display method, related device and computing equipment |
| CN111698562A (en) * | 2019-03-12 | 2020-09-22 | 青岛海尔多媒体有限公司 | Control method and device of television equipment, television equipment and storage medium |
| CN111291264B (en) * | 2020-01-23 | 2023-06-23 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207972A (en) * | 2011-06-15 | 2011-10-05 | 中山大学 | Television program recommending method and device for digital television |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090006368A1 (en) * | 2007-06-29 | 2009-01-01 | Microsoft Corporation | Automatic Video Recommendation |
| US8365227B2 (en) * | 2009-12-02 | 2013-01-29 | Nbcuniversal Media, Llc | Methods and systems for online recommendation |
| CN102088626B (en) * | 2009-12-02 | 2014-08-13 | Tcl集团股份有限公司 | On-line video recommendation method and video portal service system |
-
2013
- 2013-06-03 CN CN201310217181.9A patent/CN103324686B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207972A (en) * | 2011-06-15 | 2011-10-05 | 中山大学 | Television program recommending method and device for digital television |
Non-Patent Citations (4)
| Title |
|---|
| Exploring Folksonomy for Personalized Search;Shengliang Xu et al;《annual international ACM SIGIR conference on Research and development in information retrieval》;20081231;第155-162页 * |
| Personalized Celebrity Video Search Based on Cross-Space Mining;Zhengyu Deng et al;《13th Pacific-Rim Conference on Multimedia》;20121231;第455-463页 * |
| SocialTransfer: cross-domain transfer learning from social streams for media applications;Suman D.Roy et al;《Proceedings of the 20th ACM International Conference on Multimedia》;20121029;第649-658页 * |
| 面向网络视频推荐的协同过滤算法优化与实现;高思敏等;《网络新媒体技术》;20120331;第1卷(第2期);第22-27页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103324686A (en) | 2013-09-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103324686B (en) | Real time individual video recommendation method based on text flow network | |
| US20220174362A1 (en) | Profiling media characters | |
| Yu et al. | Robust linear regression: A review and comparison | |
| US9712588B1 (en) | Generating a stream of content for a channel | |
| US9424319B2 (en) | Social media based content selection system | |
| US10776885B2 (en) | Mutually reinforcing ranking of social media accounts and contents | |
| Stephens-Davidowitz | Google, tell me. Is my son a genius | |
| EP2800015A1 (en) | Content distribution | |
| US10437894B2 (en) | Method and system for app search engine leveraging user reviews | |
| CN103778260A (en) | Individualized microblog information recommending system and method | |
| JP2015201157A (en) | Dynamic content recommendation system using social network data | |
| US20150324342A1 (en) | Method and apparatus for enriching social media to improve personalized user experience | |
| US9036806B1 (en) | Predicting the class of future customer calls in a call center | |
| US10042944B2 (en) | Suggested keywords | |
| CN110390052A (en) | Search for recommended method, the training method of CTR prediction model, device and equipment | |
| KR101925950B1 (en) | Method and device for recommending contents based on inflow keyword and relevant keyword for contents | |
| US20220188366A1 (en) | Combined Wide And Deep Machine Learning Models For Automated Database Element Processing Systems, Methods And Apparatuses | |
| JP2013105213A (en) | Information recommending device and method, and device and program | |
| Bothe et al. | From MOOCs to micro learning activities | |
| US9020863B2 (en) | Information processing device, information processing method, and program | |
| JP6169511B2 (en) | Apparatus, program, and method for analyzing poster's psychological transition based on comment text | |
| US20160162930A1 (en) | Associating Social Comments with Individual Assets Used in a Campaign | |
| Alluhaidan | Recommender system using collaborative filtering algorithm | |
| CN110245261A (en) | A kind of latent structure method and system in multi-modal short video recommendation system | |
| CN110134773A (en) | A kind of search recommended method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |