CN109936770B - Program recommendation method based on matrix online completion and program characteristics - Google Patents
Program recommendation method based on matrix online completion and program characteristics Download PDFInfo
- Publication number
- CN109936770B CN109936770B CN201910318963.9A CN201910318963A CN109936770B CN 109936770 B CN109936770 B CN 109936770B CN 201910318963 A CN201910318963 A CN 201910318963A CN 109936770 B CN109936770 B CN 109936770B
- Authority
- CN
- China
- Prior art keywords
- program
- matrix
- completion
- programs
- correlation
- 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
Abstract
The invention discloses a program recommendation method based on matrix online completion and program characteristics, which comprises the following steps: the method comprises the following steps: s01, reading the characteristics of each program; s02, calculating the correlation among the programs; s03, solving the objective function through an iteration method to obtain a completion matrix X; and S04, selecting programs for the target audience according to the completion matrix X to recommend the programs. The scoring matrix can be adjusted in real time after the new program is put in storage or the audience scores, and the pushing pertinence and the accuracy are high. The scheme is suitable for screening and recommending the multimedia program platform.
Description
Technical Field
The invention relates to the field of media program evaluation and pushing, in particular to a program recommendation method based on matrix online completion and program characteristics.
Background
Watching/listening to media programs such as audio and video is one of the common entertainment modes of people, and with the development of science and technology and the improvement of living standard, people listen to and watch programs more on internet-based equipment such as computers, mobile phones and PADs. In order to attract users to watch a plurality of different programs and improve the audience rating, the media provider can actively push the programs to the users. How to accurately push interested programs to users is one of the important topics in the big data era. The existing recommendation algorithm is often low in efficiency and cannot meet the requirement of real-time performance. And the feature information of the program itself is not applied to the recommendation, so the recommendation accuracy is not high. Especially for the program which is just put in storage and lacks enough evaluation and playing amount, how to quickly push the program to the possibly interested group is a problem which is difficult to solve.
Disclosure of Invention
The invention mainly solves the technical problem of insufficient pushing precision in the prior art, and provides a program recommendation method based on matrix online completion and program characteristics with high precision and low evaluation requirements.
The invention mainly solves the technical problems through the following technical scheme: a program recommendation method based on matrix online completion and program characteristics comprises the following steps:
s01, reading the characteristic of each program, and recording the characteristic f of the ith programiThe feature of the jth program is fj;
S02, calculating the correlation between the programs, and marking the correlation between the ith program and the jth program as sim (f)i,fj);
S03, solving the following objective function through an iterative method:
obtaining a completion matrix X;
s04, selecting programs for the target audience to recommend according to the completion matrix X;
at time t0,t1,...,tkThe observation matrices are M respectively0,M1,…,MkThe observed index sets are respectively omega0,Ω1,…,ΩkWhereinThe kth time tkThe newly appeared value of credit isRepresents the time tkThe index set of the program corresponding to the newly appeared credit value;
||·||Fis the Frobenious norm of the matrix, λ is the regularization parameter, rank (X) is the rank of X; μ is a regularization parameter, XiuAnd XjuThe evaluation scores of the user u to the ith program and the jth program in the completion matrix are respectively.
Preferably, the kernel norm | | X | | luminance of the matrix is employed*Instead of rank (X), the objective function is
||X||*Defined as the sum of all singular values of the matrix.
Preferably, in step S03, the initial value of the iterative solution is the completion matrix X calculated at the previous timek-1。
Preferably, in step S02, the program correlation is determined by the following formula:
sim(fi,fj)=1-|fi-fj|/(fi+fj)
sim(fi,fj) Is the correlation between the ith program and the jth program.
The method has the substantial effects that the scoring matrix can be adjusted in real time after the new program is put in storage or the audience scores, the pushing pertinence is strong, and the accuracy is high.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The technical scheme of the invention is further specifically described by the following embodiments and the accompanying drawings.
Example (b): there are a large number of broadcast programs on the system platform, so recommending broadcast programs of interest to audiences is an important function of the system platform. As the audience interest changes every day, the data in the system platform is continuously updated, for example, some programs are added, some programs are scored by the audience, and so on. Therefore, the real-time performance of the recommendation has a very important influence on the accuracy of the broadcast program recommendation. The conventional recommendation algorithm performs poorly in real-time.
If the audience and the broadcast program are viewed as two dimensions, the rating of the program by the audience can be viewed as a matrix, referred to as a rating matrix. Because only a portion of the audience has scored certain programs, the elements of some positions in the scoring matrix are unknown. Recommending broadcast programs for an audience requires estimating audience preferences for the programs, i.e., predicting unknown elements in the scoring matrix, so that estimation of unknown scores can translate into a matrix-filling problem. And (3) complementing the scoring matrix under the assumption of low rank, namely, ensuring that the rank of the scoring matrix is as small as possible under the condition that the predicted scoring matrix is limited to be consistent with the existing scoring matrix on an observation set. The reason for using the low rank assumption is that there is some correlation between programs and between audiences, and therefore, the rows and columns of the scoring matrix exhibit some correlation and thus have a low dimensionality.
There are m programs, n audiences, and the observation matrix isThe required completion matrix isThe index of the element observed in M isUnder the condition of low rank limitation, the objective function of matrix completion is as follows:
whereinIs the projection operator, and the projection operator,||·||Fis the Frobenious norm of the matrix, λ is the regularization parameter, rank (X) is the rank of X. Because rank (X) is very difficult to optimize, the academic community generally employs the kernel norm of the matrix | | | X | | luminance*Instead of rank, where | | X | | non-calculation*Defined as the sum of all singular values of the matrix. Thus, the objective function of the matrix completion in practice is as follows:
the existing scoring prediction model based on low-rank matrix completion has the following defects:
(1) in practice, because the scores of audiences for programs are continuously generated, when a new group of audiences score a part of programs, the system needs to update and adjust the whole scoring matrix immediately according to the new information, so as to predict the scores of all the audiences for all the programs more accurately. Therefore, what we need to solve is actually a problem of online matrix completion.
(2) The above model does not take into account the characteristic information of the program, and only evaluates and recommends based on the audience rating. Therefore, the recommendation accuracy is not high.
To this end, consideration needs to be given to designing program recommendations based on online matrix completion and program characteristics. First, the stronger the correlation between the two programs, i.e., sim (f)i,fj) The larger the audience scores the closer the two programs. By mixingThe program recommendation method is introduced into a previous objective function as a regular term, so that the characteristics and the relevance of the program are applied to the recommendation model to improve the recommendation accuracy. The objective function thus becomes:
then, an online solving algorithm is designed for the objective function so as to meet the requirement of real-time recommendation. The main mode is that the calculation result of the last iteration is used as the initial value of the iteration in the iteration process, so that the calculation efficiency is improved.
To facilitate the description of the specific flow, we first give the following notations: set at time t0,t1,...,tkThe observation matrices are M respectively0,M1,…,MkThe observed index sets are respectively omega0,Ω1,…,ΩkWhereinThe kth time tkThe newly emerging score value is indexed by Ωk-Ωk-1. Is provided withRepresents the time tkAnd the newly appeared score value corresponds to the index set of the program.
The process is shown in figure 1:
(1) the correlation between the programs is calculated. The characteristics of each program can be obtained by using a cross-media data characteristic extraction model based on deep learning and the like, or the self-contained characteristics of the program can be directly read. Let the characteristics of the ith and jth programs be fiAnd fj。
(2) Calculating program similarity, similarity between fi and fj using sim (f)i,fj) It is shown that,
sim(fi,fj)=1-|fi-fj|/(fi+fj)
sim(fi,fj) Larger indicates that the ith and jth programs are more relevant.
(3) And solving the current subproblem. At time tkSolving the following sub-problem:
Thereby obtaining an observation matrix of MkThe scoring matrix under the conditions of (1). The problem can be solved directly by iterative methods. However, for the purpose of solving efficiency, the completion matrix obtained by the calculation at the previous moment may be used as Xk-1The iterative calculation is started for the initial value. Therefore, the blindness of initial value selection is reduced, and the convergence speed and the calculation efficiency are greatly improved.
(4) By using the model, the characteristic information of the program is applied to program recommendation, so that the recommendation accuracy is improved.
In addition, by using an online optimization algorithm, the calculation cost can be obviously reduced, and the real-time prediction of the scoring matrix is realized, so that the purpose of real-time recommendation is achieved.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Although the terms scoring matrix, completion matrix, nuclear norm, etc. are used more herein, the possibility of using other terms is not excluded. These terms are used merely to more conveniently describe and explain the nature of the present invention; they are to be construed as being without limitation to any additional limitations that may be imposed by the spirit of the present invention.
Claims (4)
1. A program recommendation method based on matrix online completion and program characteristics is characterized by comprising the following steps:
s01, reading the characteristic of each program, and recording the characteristic f of the ith programiThe feature of the jth program is fj;
S02, calculating the correlation between the programs, i-th program and j-th programThe correlation is noted as sim (f)i,fj);
S03, solving the following objective function through an iterative method:
obtaining a completion matrix X;
s04, selecting programs for the target audience to recommend according to the completion matrix X;
at time t0,t1,…,tkThe observation matrices are M respectively0,M1,…,MkThe observed index sets are respectively omega0,Ω1,…,ΩkWhereinThe kth time tkThe newly emerging score value is indexed by Ωk-Ωk-1;Represents the time tkThe index set of the program corresponding to the newly appeared credit value;
2. The matrix-based online completion and programming system of claim 1The characteristic program recommendation method is characterized in that the kernel norm of the matrix is adopted | | X | | ventilated voice*Instead of rank (X), the objective function is
||X||*Defined as the sum of all singular values of the matrix.
3. The method for recommending programs based on matrix online completion and program characteristics as claimed in claim 1 or 2, wherein in step S03, the initial value of the iterative solution is the completion matrix X calculated at the previous timek-1。
4. The method for recommending programs based on matrix online completion and program characteristics as claimed in claim 1, wherein in said step S02, the program correlation is determined by the following formula:
sim(fi,fj)=1-|fi-fj|/(fi+fj)
sim(fi,fj) Is the correlation between the ith program and the jth program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910318963.9A CN109936770B (en) | 2019-04-19 | 2019-04-19 | Program recommendation method based on matrix online completion and program characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910318963.9A CN109936770B (en) | 2019-04-19 | 2019-04-19 | Program recommendation method based on matrix online completion and program characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109936770A CN109936770A (en) | 2019-06-25 |
CN109936770B true CN109936770B (en) | 2021-04-27 |
Family
ID=66990424
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910318963.9A Active CN109936770B (en) | 2019-04-19 | 2019-04-19 | Program recommendation method based on matrix online completion and program characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109936770B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110598094A (en) * | 2019-08-23 | 2019-12-20 | 卓尔智联(武汉)研究院有限公司 | Shopping recommendation method based on matrix completion, electronic device and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101540874A (en) * | 2009-04-23 | 2009-09-23 | 中山大学 | Interactive TV program recommendation method based on collaborative filtration |
CN104965896A (en) * | 2015-06-23 | 2015-10-07 | 中国船舶重工集团公司第七0九研究所 | Recommendation method and system for representing user preference based on quadratic function |
CN105282616A (en) * | 2015-11-13 | 2016-01-27 | 云南大学 | Combined collaborative filtering IPTV program recommendation method |
CN105550240A (en) * | 2015-11-30 | 2016-05-04 | 浪潮通用软件有限公司 | Recommendation method and recommendation device |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8275764B2 (en) * | 2007-08-24 | 2012-09-25 | Google Inc. | Recommending media programs based on media program popularity |
US10515116B2 (en) * | 2016-02-08 | 2019-12-24 | Hulu, LLC | Generation of video recommendations using connection networks |
-
2019
- 2019-04-19 CN CN201910318963.9A patent/CN109936770B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101540874A (en) * | 2009-04-23 | 2009-09-23 | 中山大学 | Interactive TV program recommendation method based on collaborative filtration |
CN104965896A (en) * | 2015-06-23 | 2015-10-07 | 中国船舶重工集团公司第七0九研究所 | Recommendation method and system for representing user preference based on quadratic function |
CN105282616A (en) * | 2015-11-13 | 2016-01-27 | 云南大学 | Combined collaborative filtering IPTV program recommendation method |
CN105550240A (en) * | 2015-11-30 | 2016-05-04 | 浪潮通用软件有限公司 | Recommendation method and recommendation device |
Non-Patent Citations (1)
Title |
---|
低秩矩阵近似理论与应用;仲小伟;《中国优秀硕士学位论文全文数据库》;20160910;A002-9 * |
Also Published As
Publication number | Publication date |
---|---|
CN109936770A (en) | 2019-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10671679B2 (en) | Method and system for enhanced content recommendation | |
US9552555B1 (en) | Methods, systems, and media for recommending content items based on topics | |
US9110955B1 (en) | Systems and methods of selecting content items using latent vectors | |
US10380649B2 (en) | System and method for logistic matrix factorization of implicit feedback data, and application to media environments | |
TW202007178A (en) | Method, device, apparatus, and storage medium of generating features of user | |
CN110929052A (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
US20120143911A1 (en) | Recommendations based on topic clusters | |
US20150242750A1 (en) | Asymmetric Rankers for Vector-Based Recommendation | |
CN109511015B (en) | Multimedia resource recommendation method, device, storage medium and equipment | |
WO2014153133A1 (en) | Cross media recommendation | |
CN111966909B (en) | Video recommendation method, device, electronic equipment and computer readable storage medium | |
US20220408131A1 (en) | Image analysis system | |
CN108540860B (en) | Video recall method and device | |
CN111159563A (en) | Method, device and equipment for determining user interest point information and storage medium | |
CN115618024A (en) | Multimedia recommendation method and device and electronic equipment | |
CN109936770B (en) | Program recommendation method based on matrix online completion and program characteristics | |
WO2017101328A1 (en) | Method, device, and system for smart television to present playback content | |
US11392751B1 (en) | Artificial intelligence system for optimizing informational content presentation | |
US20240107125A1 (en) | System and method for modelling access requests to multi-channel content sharing platforms | |
CN110569447B (en) | Network resource recommendation method and device and storage medium | |
JP2007233515A (en) | Information processor and information processing method, information providing apparatus and information providing method, and program | |
CN115129922A (en) | Search term generation method, model training method, medium, device and equipment | |
CN114580790A (en) | Life cycle stage prediction and model training method, device, medium and equipment | |
CN114040216A (en) | Live broadcast room recommendation method, medium, device and computing equipment | |
CN113094584A (en) | Method and device for determining recommended learning resources |
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 | ||
CB03 | Change of inventor or designer information |
Inventor after: Shan Xuesong Inventor after: Wu Jie Inventor after: Gu Jiangbiao Inventor after: Sheng Guolin Inventor after: Zhao Fan Inventor before: Zhao Fan Inventor before: Sheng Guolin Inventor before: Gu Jiangbiao Inventor before: Shan Xuesong Inventor before: Wu Jie |
|
CB03 | Change of inventor or designer information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |