CN107635143A - The method for chasing after play on TV based on viewing behavior prediction user - Google Patents
The method for chasing after play on TV based on viewing behavior prediction user Download PDFInfo
- Publication number
- CN107635143A CN107635143A CN201711078604.8A CN201711078604A CN107635143A CN 107635143 A CN107635143 A CN 107635143A CN 201711078604 A CN201711078604 A CN 201711078604A CN 107635143 A CN107635143 A CN 107635143A
- Authority
- CN
- China
- Prior art keywords
- data
- video
- user
- viewing
- characteristic
- 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
Landscapes
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
It the present invention relates to the use of big data and carry out user's behavior prediction technology, it discloses a kind of method for chasing after play on TV based on viewing behavior prediction user, so as to shorten the time that user finds preference video, improves the efficiency of start viewing video, lifts Consumer's Experience.This method comprises the following steps:A. feature extraction is carried out to the history viewing behavioral data of TV based on user;B. the data creating data set of extraction is utilized;C. the data set of making is inputted to Logic Regression Models and is trained and verifies;D. it is predicted using the viewing behavior of training and the Logic Regression Models verified to user.
Description
Technical field
It the present invention relates to the use of big data and carry out user's behavior prediction technology, and in particular to one kind is based on viewing behavior prediction
The method that user chases after play on TV.
Background technology
With the development of big data, substantial amounts of user data is have accumulated in terminal device manufacturer hand, how according to these use
User data come improve the Consumer's Experience of product be instantly major manufacturer terminal all in the thing done.
Intelligent television is one of large-size screen monitors of internet three, and the function that TV provides at present is no longer the straight of viewing TV station already
Broadcast so single, various program sources, various applications can be watched and used on TV.The selection of user is more diversified, its
The personalization of viewing behavior is just more obvious.
But for the intelligent television of service with no personalization, start will all pass through same process, into same shape
State, user must find the video oneself liked or application manually.And in fact, user behavior is typically to have certain rule
, these rules are all lain among the behavioral data that it uses TV.
If the usage behavior data of user, the behavior of Accurate Prediction user next time so that after start just can be passed through
Jump directly in the content to be watched of user, it is possible to shorten the time that user finds favor program;Or when user watches
During other video, user is reminded, its video " chased after " has been updated over, it is possible to increase user watches efficiency, strengthens user couple
The dependence feeling of TV.
The content of the invention
The technical problems to be solved by the invention are:It is proposed it is a kind of based on viewing behavior prediction user chase after on TV play
Method, so as to shorten the time that user finds preference video, the efficiency of start viewing video is improved, lifts Consumer's Experience.
The present invention solves the technical scheme that above-mentioned technical problem uses:
The method for being chased after play on TV based on viewing behavior prediction user, is comprised the following steps:
A. feature extraction is carried out to the history viewing behavioral data of TV based on user;
B. the data creating data set of extraction is utilized;
C. the data set of making is inputted to Logic Regression Models and is trained and verifies;
D. it is predicted using the viewing behavior of training and the Logic Regression Models verified to user.
Optimize as further, in step d, in addition to:If predicting user is chasing after play, directly jumped after being started shooting in next time
Go to the newest viewing progress progress video playback that user chases after play.
Optimize as further, in step a, the user includes to the history viewing behavioral data of TV:
Video playback data:Enter the time shaft of video player including user, open and exit the time of certain video
Axle;
TV turning on-off data:Including television startup time shaft and unused time axle;
Matchmaker's standing ceases:For the video information of broadcasting, including the video name of broadcasting, video ID, video sequence ID, video class
Type and brief introduction.
Optimize as further, in step a, the feature extraction, specifically include:
A1. the unlatching of television startup, unused time axle and video, post-set time axle are changed into standard time stamp, and on time
Between stab ascending sort, be integrated into the data according to time series;
A2. invalid data is cleaned, the invalid data includes:Available machine time is too short, video playback time is too short
Data;
A3. according to the data after cleaning, the feature that can be calculated is extracted;
A4. data are carried out with expectationization processing, is 0 by the processing of data desired value;
A5. principal character is chosen from the data after expectationization processing using PCA, and carries out combinations of features.
Optimize as further, in step a3, the feature that can be calculated includes:Pass through switching on and shutting down time shaft meter
Calculate start duration, video opening time axle calculates the number watched first after video is started shooting;Open and exit according to video
Time shaft calculates video-see duration;According to the family ID of video is identical, video ID differences calculate the collection number of viewing video, with
And portion's number of video is watched daily.
Optimize as further, it is described that data are carried out with expectationization processing in step a4, it is 0 by the processing of data desired value
Specifically include:
From the characteristic of step a3 extractions, a number of characteristic is extracted at random by row, by what is extracted
Characteristic seeks the desired value of each row by unit is classified as, then by this arrange in all elements subtract desired value, operation is completed
Afterwards, the expectation of all row is all 0, then whole characteristic matrix is desired for 0.
Optimize as further, in step a5, chosen and led from the data after expectationization processing using PCA
Feature is wanted, and carries out combinations of features, is specifically included:The characteristic value row of matrix are obtained after feature decomposition is done to characteristic matrix
Table, each characteristic value correspond to one-dimensional characteristic vector, and characteristic value is bigger, and the dimensional feature vector is more important, conversely, characteristic value is smaller,
The dimensional feature vector is more secondary, the maximum n dimensional features of characteristic value is taken, as principal character, then according to the principal character of selection
Between linear relationship carry out combinations of features, so that it is determined that final characteristic dimension.
Optimize as further, in step b, using the data creating data set of extraction, specifically include:
B1. all data of all TVs within the observation period are traveled through, every of each television-viewing total collection number is big
Listed in all main features of the video of 10 collection, and countershaft ascending order arranges on time;
B2 is more than 3 times according to viewing number, and viewing collection number is more than 5 collection, and start next time have viewed the video and be defined as chasing after
Play, label is made, if label value is 1, shows to start shooting next time and continue to have viewed the video, if label value is 0, shown next time
The video is not watched in start, and characteristic is corresponded to label data and associated;
B3. characteristic is normalized:The maximum of every dimensional feature is obtained, the equal divided by maximum per dimensional feature;
B4. the proper polynomial expansion module in sklearn storehouses is called, characteristic is subjected to Polynomial Expansion, is extended
Into the multistage and characteristic set that is mutually related;
B5. the training dataset and validation data set of sky are built, then by all characteristics and label data according to 7:1
Ratio assign at random in training dataset and validation data set.
B6. respectively by training characteristics data, training label data, checking characteristic, the corresponding output of checking label data
Into 4 texts.
Optimize as further, in step c, described input the data set of making to Logic Regression Models is trained
When, add second order norm canonical in the Logic Regression Models, second order norm canonical is by the parameter in model along characteristic
According to Hessian matrixes in each characteristic vector direction zoom in and out, feature is main, and scaling is bigger, and feature is more secondary, contracting
It is smaller to put ratio.
The beneficial effects of the invention are as follows:The present invention is that terminal is real by the video-see behavior of high in the clouds intelligent predicting user
Existing personalized start provides Data safeguard strong in real time.Simultaneously, it is also possible to make the personalized recommendation of video, started shooting in user
When, play out on the video for the preference for jumping directly to user, improve the efficiency that video playback is searched in start, lift user's body
Test.
Brief description of the drawings
Fig. 1 is the flow chart for establishing forecast model of the present invention;
Fig. 2 is characterized the flow chart of extraction;
Fig. 3 is data set Making programme figure.
Embodiment
The present invention is directed to propose a kind of method for chasing after play on TV based on viewing behavior prediction user, so as to shorten user
The time of preference video is found, the efficiency of start viewing video is improved, lifts Consumer's Experience.The present invention is by from terminal big data
The each user of middle extraction watches the feature of video, for example, watch the acute number in same portion, continuously watch number of days, viewing collection number, when
It is long etc., Feature Engineering is established, forecast model is established by the method for machine learning, judges whether user chases after by forecast model
It is acute.
The step of wherein establishing forecast model is as shown in figure 1, it includes:Feature extraction, data set make, the training of model
With three parts of checking, the model for being predicted to user behavior is ultimately generated:
1st, feature extraction:
Feature extraction is carried out to the history viewing behavioral data of TV based on user;The present invention makes in specific implementation
History viewing behavioral data is as shown in the table:
Because the data that can be taken are not image, this abstract data of text, but the daily record data of user's operation, example
Time shaft, the time shaft of unlatching video such as start is this.User's operation behavior is that timesharing is carried out, can be by data time sequence
Rowization, then again from the extracting data feature after serializing.
N number of TV viewing behavioral data of 10 days is extracted because former data volume is too big, therefore using random fashion, will be opened
Machine, video open, video exit, this 4 kinds of behaviors of shutting down according to time series, some nothings in the data after serializing be present
Imitate data, in order to improve data-handling efficiency, avoid the waste of invalid data processing time, can the available machine time is too short, regard
Frequency plays the data cleansing such as too short and fallen, in specific implementation, by setting threshold value, such as:Available machine time is less than 10 minutes
Wash, primary video reproduction time washing less than 5 minutes.
Data after over cleaning can calculate user and watch the multidimensional characteristics such as the duration of video, collection number, viewing number of days
Data.For these features which and chase after acute strong correlation, which is weak related or unrelated, can not intuitively obtain.Therefore, to carrying
All features got will be changed using principal component analytical method (PCA) and unconspicuous feature weeds out.
The principle of principal component analytical method is to obtain the feature value list of matrix, feature after feature decomposition is done to eigenmatrix
Matrix is equal to the covariance matrix of whole input feature vector data, and each characteristic value corresponds to one-dimensional characteristic vector, and characteristic value is bigger,
The dimensional feature vector is more important, conversely, characteristic value is smaller, the dimensional feature vector is more secondary.The maximum n Wei Te of characteristic value can be taken
Sign, as principal character.
2nd, the making of data set:
This part is the data set for utilizing the data creating of extraction to be used for data training and data verification, in order in next step
Carry out the training and checking of model.It should be noted that in this part except making data training using the characteristic of extraction
Outside the data set of data verification, the making of label data is also related to, so that characteristic to be marked, label is 1 table
Show when meeting certain condition (such as:Watch number and/or viewing collection number etc.) start next time still watching play, mark
Sign and represent to start shooting next time for 0 not watching the play;Label data is associated with characteristic, is easy to the training of characteristic
And checking.
3rd, the training and checking of model:
The present invention is using logistic regression as forecast model.And use the fitting of in general logistic regression training characteristics data
Poor effect.Then proper polynomial extension is introduced to characteristic.
Proper polynomial n ranks extend the 1-n rank powers that three-dimensional feature (a, b, c) can be extended to each element, and mutually
Combination.For example, (a, b, c) does the extension of 2 ranks, it is extended to (1, a, a*a, b, b*b, c, c*c, ab, ac, bc).With the spy after extension
Data are levied to train logical model.
In order to prevent over-fitting, the present invention returns logical model and adds a second order norm canonical, second order norm canonical
Have the ability of each characteristic vector direction scaling in Hessian matrixes of the parameter in model along characteristic, feature is more main
Will, scaling is bigger, and feature is more secondary, and scaling is smaller.Therefore, so can be by the influence of the secondary feature after extension
Very little is narrowed down to, and the size of principal character is kept approximately constant.Made using the Logic Regression Models after training and checking
For last forecast model, the behavior of user can be predicted, to provide personalized start service.
Embodiment:
The method for being chased after play in the present embodiment on TV based on viewing behavior prediction user, is comprised the following steps:
A. feature extraction is carried out to the history viewing behavioral data of TV based on user;
Feature extraction flow in this step is as shown in Fig. 2 it includes:
A1. data time series:
Start, unused time axle and video are opened, post-set time axle changes into standard time stamp, and temporally stabs ascending order
Sequence, start data and video data are integrated into the data according to time series together.
A2. data cleansing:
Effectively start is just calculated once according at least watching 10 minutes after start, 5 points are at least watched according to a video is opened
Clock calculates the effectively viewing of an effective video to clear up data, and is that unit (cover id are unit) arranges according to " portion " by video
Data.
A3. whole feature extractions:
According to the data after data cleansing, all features that can be calculated all are extracted.For example, pass through switch
Machine time shaft calculates start duration, video opening time axle calculates the number watched first after video is started shooting;Opened according to video
The time shaft for opening and exiting calculates video-see duration;Viewing is calculated according to the cover id of video are identical, video id are different
The collection number of video, and portion's number of viewing video etc. daily.
Generally speaking, all features such as following table that can be extracted, all features all see a video according to a television terminal
For index:
Feature | Feature number |
(video) continuously starts shooting first by viewing number | ① |
(video) watches number of days | ② |
" portion " number of (same day for watching the video) viewing video | ③ |
(video) is by viewing number | ④ |
(video) has watched collection number | ⑤ |
(video) has watched total duration | ⑥ |
(video) still remaining collection number | ⑦ |
Total collection number of (video) | ⑧ |
A4. data it is expected to turn to 0:
From all characteristics, by row 1/10th (former data volume is larger) of random extraction, the feature that will be extracted
Data seek the desired value of each row by unit is classified as, then by this arrange in all elements subtract desired value.After the completion of operation, institute
The expectation for having row is all 0, and the expectation of whole characteristic matrix is also just 0.
A5. principal character is extracted by principal component analysis, and carries out combinations of features:
It will it is expected that turning to the data set after 0 uses principal component analytical method, leaves 5 dimension significant datas.Wherein, PCA methods
According to general principle, the covariance class and feature decomposition class completion program in numpy storehouses are called by python, then data set
It is input to after matrixing in algorithm, seeks five dimension datas that wherein characteristic value is maximum, the feature of its object seeks to the main of reservation
Feature.
By computing, 3., 4., 5., 6., 8. the feature remained is
In common prudence, 5., 6. this two dimensional feature chase after play in most cases (it is fast that a minority watches certain rally
Enter) it is linearly related, i.e. y=ax, wherein, y is viewing duration, and x is collection number, and a is often to collect duration.If it is implicitly present in this
It is linearly related, then this two dimensional feature selects 1 dimension can, and collection number is independent variable, and total duration is dependent variable, therefore, selected works number.
The distribution of the collection number and duration of real data can be seen that collection number and duration is implicitly present in such linear relationship, and simply a is not
One value, but a span.Here 6. number feature is still removed.
Because 5. and 8. feature is all collection number, one is viewing collection number, and one is total collection number, 5. divided by is 8. combined into viewing
The progress of video, meanwhile, the condition that this feature is set up is total collection number at least above 10 collection.
So, whole Feature Engineering includes 3-dimensional feature:Watch number, viewing progress, the general headquarters of same day viewing video
Number.
B. the data creating data set of extraction is utilized;
Made according to the characteristic of the principal character dimension that is finally determined in step a and extraction for training and verifying
Data set, Making programme is as shown in figure 3, it comprises the following steps:
B1. all data of all TVs within the observation period are traveled through, every of each television-viewing total collection number is big
Listed in all main features of the video of 10 collection, and countershaft ascending order arranges on time;
B2. it is more than 3 times according to viewing number, viewing collection number is more than 5 collection, and start next time have viewed the video and be defined as chasing after
Play, label is made, if label value is 1, shows to start shooting next time and continue to have viewed the video, if label value is 0, shown next time
The video is not watched in start, and characteristic is corresponded to label data and associated;
B3. characteristic is normalized:The maximum of every dimensional feature is obtained, the equal divided by maximum per dimensional feature;
B4. the proper polynomial expansion module in sklearn storehouses is called, characteristic is subjected to Polynomial Expansion, is extended
Into the multistage and characteristic set that is mutually related;
B5. the training dataset and validation data set of sky are built, then by all characteristics and label data according to 7:1
Ratio assign at random in training dataset and validation data set.
B6. respectively by training characteristics data, training label data, checking characteristic, the corresponding output of checking label data
Into 4 texts.
C. the data set of making is inputted to Logic Regression Models and is trained and verifies;
Logic Regression Models are write with the LogisticRegression classes in python language calls sklearn, are selected
It is optimization method that stochastic averagina gradient, which declines ' sag ', to avoid overfitting, selects second order norm canonical (L2 canonicals) to train
It is trained in data-in logic regression model in data set (including training characteristics data and training label data);It will test
Verified, made in data-in logic regression model in card data set (including checking characteristic and checking label data)
Its precision is up to standard.
D. it is predicted using the viewing behavior of training and the Logic Regression Models verified to user.
Logic Regression Models after constantly training and checking are substantially at stable state, can be as prediction mould
Viewing behavior of the type to user is predicted, so as to provide the user the start service of personalization or personalized ventilation system.
Claims (9)
1. chase after the method for play on TV based on viewing behavior prediction user, it is characterised in that comprise the following steps:
A. feature extraction is carried out to the history viewing behavioral data of TV based on user;
B. the data creating data set of extraction is utilized;
C. the data set of making is inputted to Logic Regression Models and is trained and verifies;
D. it is predicted using the viewing behavior of training and the Logic Regression Models verified to user.
2. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 1, it is characterised in that step
In d, in addition to:If predicting user is chasing after play, jumped directly to after being started shooting in next time user chase after it is acute it is newest watch into
Degree carries out video playback.
3. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 1, it is characterised in that step
In a, the user includes to the history viewing behavioral data of TV:
Video playback data:Enter the time shaft of video player including user, open and exit the time shaft of certain video;
TV turning on-off data:Including television startup time shaft and unused time axle;
Matchmaker's standing ceases:For the video information of broadcasting, including the video name of broadcasting, video ID, video sequence ID, video type and
Brief introduction.
4. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 3, it is characterised in that step
In a, the feature extraction, specifically include:
A1. television startup, unused time axle and video are opened, post-set time axle changes into standard time stamp, and temporally stabbed
Ascending sort, it is integrated into the data according to time series;
A2. invalid data is cleaned, the invalid data includes:The number that available machine time is too short, video playback time is too short
According to;
A3. according to the data after cleaning, the feature that can be calculated is extracted;
A4. data are carried out with expectationization processing, is 0 by the processing of data desired value;
A5. principal character is chosen from the data after expectationization processing using PCA, and carries out combinations of features.
5. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 4, it is characterised in that step
In a3, the feature that can be calculated includes:Start duration, video opening time axle meter are calculated by switching on and shutting down time shaft
Calculate the number watched first after video is started shooting;The time shaft opened and exited according to video calculates video-see duration;According to
The family ID of video is identical, video ID differences calculate the collection number of viewing video, and portion's number of viewing video daily.
6. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 4, it is characterised in that step
It is described that data are carried out with expectationization processing in a4, the processing of data desired value is specifically included for 0:
From the characteristic of step a3 extractions, a number of characteristic is extracted at random by row, the feature that will be extracted
Data seek the desired value of each row by unit is classified as, then by this arrange in all elements subtract desired value, after the completion of operation, institute
To have the expectations of row be all 0, then whole characteristic matrix is desired for 0.
7. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 4, it is characterised in that step
In a5, principal character is chosen from the data after expectationization processing using PCA, and carries out combinations of features, specific bag
Include:The feature value list of matrix is obtained after feature decomposition is done to characteristic matrix, each characteristic value corresponds to one-dimensional characteristic vector,
Characteristic value is bigger, and the dimensional feature vector is more important, conversely, characteristic value is smaller, the dimensional feature vector is more secondary, takes characteristic value maximum
N dimensional features, as principal character, combinations of features is then carried out according to the linear relationship between the principal character of selection,
So that it is determined that final characteristic dimension.
8. chase after the method for play on TV based on viewing behavior prediction user as claimed in claim 7, it is characterised in that step
In b, using the data creating data set of extraction, specifically include:
B1. all data of all TVs within the observation period are traveled through, every of each television-viewing total collection number is more than 10
All main features of the video of collection are listed, and countershaft ascending order arranges on time;
B2 is more than 3 times according to viewing number, and viewing collection number is more than 5 collection, and start next time have viewed the video and be defined as chasing after play, make
Make label, if label value is 1, shows to start shooting next time and continue to have viewed the video, if label value is 0, show to start shooting next time
The video is not watched, and characteristic is corresponded to label data and associated;
B3. characteristic is normalized:The maximum of every dimensional feature is obtained, the equal divided by maximum per dimensional feature;
B4. the proper polynomial expansion module in sklearn storehouses is called, characteristic is subjected to Polynomial Expansion, is extended to more
Rank and the characteristic set that is mutually related;
B5. the training dataset and validation data set of sky are built, then by all characteristics and label data according to 7:1 ratio
Example is assigned in training dataset and validation data set at random;
B6. training characteristics data, training label data, checking characteristic, checking label data are correspondingly output to 4 texts respectively
In this document.
9. the method for chasing after play on TV based on viewing behavior prediction user as described in claim 1-8 any one, it is special
Sign is, in step c, it is described by the data set of making input to Logic Regression Models be trained when, in the logistic regression
Model adds second order norm canonical.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711078604.8A CN107635143B (en) | 2017-11-06 | 2017-11-06 | Method for predicting user's drama chase on television based on watching behavior |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711078604.8A CN107635143B (en) | 2017-11-06 | 2017-11-06 | Method for predicting user's drama chase on television based on watching behavior |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107635143A true CN107635143A (en) | 2018-01-26 |
CN107635143B CN107635143B (en) | 2020-05-05 |
Family
ID=61106802
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711078604.8A Active CN107635143B (en) | 2017-11-06 | 2017-11-06 | Method for predicting user's drama chase on television based on watching behavior |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107635143B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108628983A (en) * | 2018-04-27 | 2018-10-09 | 正方软件股份有限公司 | A kind of data processing method and device |
CN109511015A (en) * | 2018-08-10 | 2019-03-22 | 腾讯科技(深圳)有限公司 | Multimedia resource recommended method, device, storage medium and equipment |
CN110149540A (en) * | 2018-04-27 | 2019-08-20 | 腾讯科技(深圳)有限公司 | Recommendation process method, apparatus, terminal and the readable medium of multimedia resource |
CN116156263A (en) * | 2023-03-06 | 2023-05-23 | 四川长虹电器股份有限公司 | Real-time user chasing processing method |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103914743A (en) * | 2014-04-21 | 2014-07-09 | 中国科学技术大学先进技术研究院 | On-line serial content popularity prediction method based on autoregressive model |
CN104834657A (en) * | 2014-08-27 | 2015-08-12 | 腾讯科技(北京)有限公司 | User behavior analysis method and servers |
CN105095431A (en) * | 2015-07-22 | 2015-11-25 | 百度在线网络技术(北京)有限公司 | Method and device for pushing videos based on behavior information of user |
CN204859436U (en) * | 2015-04-07 | 2015-12-09 | 广东省广播电视网络股份有限公司中山分公司 | Broadcasting digital television's user experience system |
US20160044357A1 (en) * | 2014-08-08 | 2016-02-11 | TCL Research America Inc. | Personalized channel recommendation method and system |
CN105468770A (en) * | 2015-12-09 | 2016-04-06 | 合一网络技术(北京)有限公司 | Data processing method and system |
CN106372113A (en) * | 2016-08-22 | 2017-02-01 | 上海亿账通互联网科技有限公司 | News content pushing method and system |
CN106776873A (en) * | 2016-11-29 | 2017-05-31 | 珠海市魅族科技有限公司 | A kind of recommendation results generation method and device |
-
2017
- 2017-11-06 CN CN201711078604.8A patent/CN107635143B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103914743A (en) * | 2014-04-21 | 2014-07-09 | 中国科学技术大学先进技术研究院 | On-line serial content popularity prediction method based on autoregressive model |
US20160044357A1 (en) * | 2014-08-08 | 2016-02-11 | TCL Research America Inc. | Personalized channel recommendation method and system |
CN104834657A (en) * | 2014-08-27 | 2015-08-12 | 腾讯科技(北京)有限公司 | User behavior analysis method and servers |
CN204859436U (en) * | 2015-04-07 | 2015-12-09 | 广东省广播电视网络股份有限公司中山分公司 | Broadcasting digital television's user experience system |
CN105095431A (en) * | 2015-07-22 | 2015-11-25 | 百度在线网络技术(北京)有限公司 | Method and device for pushing videos based on behavior information of user |
CN105468770A (en) * | 2015-12-09 | 2016-04-06 | 合一网络技术(北京)有限公司 | Data processing method and system |
CN106372113A (en) * | 2016-08-22 | 2017-02-01 | 上海亿账通互联网科技有限公司 | News content pushing method and system |
CN106776873A (en) * | 2016-11-29 | 2017-05-31 | 珠海市魅族科技有限公司 | A kind of recommendation results generation method and device |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108628983A (en) * | 2018-04-27 | 2018-10-09 | 正方软件股份有限公司 | A kind of data processing method and device |
CN110149540A (en) * | 2018-04-27 | 2019-08-20 | 腾讯科技(深圳)有限公司 | Recommendation process method, apparatus, terminal and the readable medium of multimedia resource |
CN110149540B (en) * | 2018-04-27 | 2021-08-24 | 腾讯科技(深圳)有限公司 | Recommendation processing method and device for multimedia resources, terminal and readable medium |
CN109511015A (en) * | 2018-08-10 | 2019-03-22 | 腾讯科技(深圳)有限公司 | Multimedia resource recommended method, device, storage medium and equipment |
CN109511015B (en) * | 2018-08-10 | 2021-12-14 | 腾讯科技(深圳)有限公司 | Multimedia resource recommendation method, device, storage medium and equipment |
CN116156263A (en) * | 2023-03-06 | 2023-05-23 | 四川长虹电器股份有限公司 | Real-time user chasing processing method |
Also Published As
Publication number | Publication date |
---|---|
CN107635143B (en) | 2020-05-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021042826A1 (en) | Video playback completeness prediction method and apparatus | |
CN107635143A (en) | The method for chasing after play on TV based on viewing behavior prediction user | |
US11636335B2 (en) | System and method for content discovery | |
Pangarker et al. | The determinants of box office performance in the film industry revisited | |
CN106131601A (en) | Video recommendation method and device | |
CA3150500C (en) | Uploader matching method and device | |
CN106294830A (en) | The recommendation method and device of multimedia resource | |
Adomavicius et al. | Recommender systems, consumer preferences, and anchoring effects | |
CN105095431A (en) | Method and device for pushing videos based on behavior information of user | |
CN106294783A (en) | A kind of video recommendation method and device | |
CN105847985A (en) | Video recommendation method and device | |
CN105430505B (en) | A kind of IPTV program commending methods based on combined strategy | |
CN109511015A (en) | Multimedia resource recommended method, device, storage medium and equipment | |
CN103559206A (en) | Information recommending method and system | |
JP2008542870A (en) | Method and apparatus for estimating the overall interest of a group of users for content | |
CN106469170B (en) | The treating method and apparatus of text data | |
US20230353828A1 (en) | Model-based data processing method and apparatus | |
EP2899989A1 (en) | Data analysis apparatus, method and program | |
CN109996122A (en) | A kind of video recommendation method, device, server and storage medium | |
CN112507163A (en) | Duration prediction model training method, recommendation method, device, equipment and medium | |
Knapp et al. | Does 3D make sense for Hollywood? The economic implications of adding a third dimension to hedonic media products | |
CN112333545B (en) | Television content recommendation method, system, storage medium and smart television | |
CN113852867B (en) | Method and device for recommending programs based on kernel density estimation | |
CN105072495B (en) | Count method, the method for pushing and device of program of personage's temperature | |
CN111669654B (en) | Program recommendation method and device, electronic equipment and readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |