CN107635143B - Method for predicting user's drama chase on television based on watching behavior - Google Patents
Method for predicting user's drama chase on television based on watching behavior Download PDFInfo
- Publication number
- CN107635143B CN107635143B CN201711078604.8A CN201711078604A CN107635143B CN 107635143 B CN107635143 B CN 107635143B CN 201711078604 A CN201711078604 A CN 201711078604A CN 107635143 B CN107635143 B CN 107635143B
- Authority
- CN
- China
- Prior art keywords
- data
- video
- characteristic
- user
- television
- 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
Landscapes
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a technology for predicting user behaviors by utilizing big data, and discloses a method for predicting a series traced on a television by a user based on watching behaviors, so that the time for the user to search for a preferred video is shortened, the efficiency of watching the video when the television is started is improved, and the user experience is improved. The method comprises the following steps: a. extracting characteristics based on historical watching behavior data of a television by a user; b. making a data set by using the extracted data; c. inputting the manufactured data set into a logistic regression model for training and verification; d. and predicting the viewing behavior of the user by using the trained and verified logistic regression model.
Description
Technical Field
The invention relates to a technology for predicting user behavior by utilizing big data, in particular to a method for predicting user drama chase on a television based on watching behavior.
Background
With the development of big data, a great amount of user data is accumulated in terminal equipment manufacturers, and how to improve the user experience of products according to the user data is what the big terminal manufacturers are doing at present.
The intelligent television is one of three screens of the internet, the function provided by the television is no longer single as that of watching the live broadcast of a television station, and various program sources and various applications can be watched and used on the television. The user's selection is more diversified, and the personalization of his viewing behavior is more obvious.
However, for the smart television without the personalized service, the same process is carried out to enter the same state when the smart television is turned on, and the user must manually search for favorite videos or applications. In fact, the user behavior is usually regular, and these rules are implicit in the behavior data of the user using the television.
If the next behavior of the user can be accurately predicted through the using behavior data of the user, the user can directly jump to the content to be watched by the user after starting the computer, and the time for searching the favorite program by the user can be shortened; or when the user watches other videos, the user is reminded that the videos which are tracking the user are updated, so that the watching efficiency of the user can be improved, and the dependence of the user on the television is enhanced.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the method for predicting the user's drama tracing on the television based on the watching behavior is provided, so that the time for the user to find the preferred video is shortened, the efficiency of watching the video when the television is started is improved, and the user experience is improved.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method for predicting a user's drama chase on a television based on viewing behavior, comprising the steps of:
a. extracting characteristics based on historical watching behavior data of a television by a user;
b. making a data set by using the extracted data;
c. inputting the manufactured data set into a logistic regression model for training and verification;
d. and predicting the viewing behavior of the user by using the trained and verified logistic regression model.
As a further optimization, step d further includes: and if the situation that the user is in the episode chase is predicted, directly jumping to the latest watching progress of the episode traced by the user for video playing after the next startup.
As a further optimization, in step a, the historical viewing behavior data of the television by the user includes:
video playing data: the method comprises the steps of a time axis of a user entering a video player and a time axis of starting and exiting a certain video;
television on-off data: the television comprises a television startup time axis and a television shutdown time axis;
media information: the video information to be played comprises the name of the played video, the video ID, the video series ID, the video type and the brief introduction.
As a further optimization, in step a, the feature extraction specifically includes:
a1. converting a television startup and shutdown time axis and a video startup and exit time axis into standard time stamps, sequencing the time stamps in an ascending order, and integrating the time stamps into data according to time serialization;
a2. flushing invalid data, the invalid data comprising: data with too short starting time and video playing time;
a3. extracting characteristics which can be calculated according to the cleaned data;
a4. carrying out expectation processing on the data, and processing the expected value of the data into 0;
a5. and selecting main characteristics from the data subjected to the anticipatory processing by adopting a principal component analysis method, and combining the characteristics.
As a further optimization, in step a3, the characteristics that can be calculated include: calculating the starting-up time length through a starting-up and closing-down time axis, and calculating the first watching times of the video after the video is started up through a video starting time axis; calculating the video watching duration according to the time axes of video starting and exiting; the number of sets of videos to be watched and the number of sections of videos to be watched per day are calculated based on the fact that the series IDs of videos are the same and the video IDs are different.
As a further optimization, in step a4, the performing expectation processing on the data, where the processing the data expectation value to 0 specifically includes:
and c, randomly extracting a certain amount of feature data from the feature data extracted in the step a3 according to rows, calculating an expected value of each column according to the extracted feature data by column units, subtracting the expected values from all elements in the column, and after the operation is finished, the expectation of all the columns is 0, so that the expectation of the whole feature data matrix is 0.
As a further optimization, in step a5, a principal component analysis method is used to select main features from the data after the expectation processing, and the feature combination is performed, which specifically includes: and performing characteristic decomposition on the characteristic data matrix to obtain a characteristic value list of the matrix, wherein each characteristic value corresponds to a one-dimensional characteristic vector, the larger the characteristic value is, the more important the dimensional characteristic vector is, otherwise, the smaller the characteristic value is, the less important the dimensional characteristic vector is, the n-dimensional characteristic with the maximum characteristic value is taken as a main characteristic, and then, characteristic combination is performed according to the linear correlation relationship among the selected main characteristics, so that the final characteristic dimension is determined.
As a further optimization, in step b, the data set is produced by using the extracted data, specifically including:
b1. traversing all data of all televisions in an observation time period, listing all main characteristics of videos with the total collection number of each television watched being more than 10 sets, and arranging the main characteristics in ascending order according to a time axis;
b2, according to the condition that the watching times are more than 3 times, the watching collection number is more than 5 collections, the video is watched at the next starting, the video is defined as a series, a label is made, if the label value is 1, the video is continuously watched at the next starting, if the label value is 0, the video is not watched at the next starting, and the characteristic data and the label data are correspondingly associated;
b3. normalizing the characteristic data: solving the maximum value of each dimension of features, and dividing each dimension of features by the maximum value;
b4. calling a characteristic polynomial expansion module in a sklern library, and performing polynomial expansion on the characteristic data to expand the characteristic data into a multi-order and mutually associated characteristic set;
b5. and constructing an empty training data set and a verification data set, and then randomly distributing all feature data and label data into the training data set and the verification data set according to the proportion of 7: 1.
b6. And correspondingly outputting the training characteristic data, the training label data, the verification characteristic data and the verification label data to a 4-text file.
As a further optimization, in step c, when the created data set is input to a logistic regression model for training, a second-order norm is added to the logistic regression model, and the second-order norm normalizes parameters in the model along the directions of the feature vectors in the Hessian matrix of the feature data, wherein the more dominant the features are, the larger the scaling is, the less subordinate the features are, and the smaller the scaling is.
The invention has the beneficial effects that: according to the invention, the video watching behavior of the user is intelligently predicted through the cloud, so that real-time powerful data guarantee is provided for realizing personalized startup of the terminal. Meanwhile, the method can be used for personalized recommendation of videos, when the user starts the computer, the user directly skips to the video preferred by the user to play, the efficiency of searching for the video when starting the computer is improved, and the user experience is improved.
Drawings
FIG. 1 is a flow chart of the present invention for building a predictive model;
FIG. 2 is a flow chart of feature extraction;
FIG. 3 is a data set production flow diagram.
Detailed Description
The invention aims to provide a method for predicting the drama chase of a user on a television based on watching behaviors, so that the time for the user to search for a preferred video is shortened, the efficiency of watching the video when the television is started is improved, and the user experience is improved. The method comprises the steps of extracting the characteristics of videos watched by each user from terminal big data, such as the number of times of watching the same drama, the number of continuous watching days, the number of watching sets, the duration and the like, establishing characteristic engineering, establishing a prediction model by a machine learning method, and judging whether the user pursues the drama or not by the prediction model.
The steps of establishing the prediction model are shown in fig. 1, and include: the method comprises three parts of feature extraction, data set making, model training and verification, and finally generates a model for predicting user behaviors:
1. feature extraction:
the method comprises the steps that feature extraction is carried out on the basis of historical watching behavior data of a television of a user; in a specific implementation of the present invention, the used historical viewing behavior data is shown in the following table:
the data that can be accessed is not abstract data such as images and texts, but log data of user operation, such as a time axis of starting up and a time axis of starting up videos. The user operation behavior is carried out in a time-sharing mode, data can be subjected to time serialization, and then features are extracted from the serialized data.
Because the original data volume is too big, therefore adopt the viewing action data of N televisions 10 days of random mode extraction, 4 kinds of actions of starting, video opening, video withdraw from, shut down are according to the time serialization, there are some invalid data in the data after the serialization, in order to improve data processing efficiency, avoid the waste of invalid data processing time, can wash data such as the start-up time is too short, the video broadcast is too short, when specifically realizing, through setting up the threshold value, if: and one-time startup time is less than 10 minutes of cleaning, and one-time video playing time is less than 5 minutes of cleaning.
The cleaned data can calculate the multi-dimensional characteristic data such as the video watching time length, the video watching collection number, the video watching days and the like of the user. Which of these features are strongly related to chase and which are weakly related or unrelated are not intuitively accessible. Therefore, the Principal Component Analysis (PCA) method is used to eliminate the features with less obvious variation from all the extracted features.
The principle of the principal component analysis method is that a characteristic matrix is subjected to characteristic decomposition to obtain a characteristic value list of the matrix, the characteristic matrix is equal to a covariance matrix of the whole input characteristic data, each characteristic value corresponds to a one-dimensional characteristic vector, the more the characteristic value is, the more important the dimensional characteristic vector is, otherwise, the smaller the characteristic value is, the more important the dimensional characteristic vector is. The n-dimensional feature with the largest feature value may be taken as the main feature.
2. And (3) making a data set:
the part is to use the extracted data to make a data set for data training and data verification so as to facilitate the next step of model training and verification. It should be noted that, in this section, in addition to making a data set for data training and data verification by using the extracted feature data, the data set also relates to making label data to mark the feature data, where a label of 1 indicates that the play is still being watched at the next startup when a certain condition is met (for example, the number of viewing times and/or the number of viewing sets, etc.), and a label of 0 indicates that the play is not being watched at the next startup; and the label data is associated with the characteristic data, so that the training and verification of the characteristic data are facilitated.
3. Training and verifying the model:
the invention uses logistic regression as a prediction model. And the fitting effect of training the feature data by adopting general logistic regression is poor. A feature polynomial extension is then introduced to the feature data.
The feature polynomial n-th order expansion can expand the three-dimensional features (a, b, c) to the 1-nth order powers of each element and combine them with each other. For example, (a, b, c) is extended by 2 steps to (1, a, b, c, ab, ac, bc). The logic model is trained with the extended feature data.
In order to prevent overfitting, a second-order norm regular is added to the logic model, the second-order norm regular has the capability of scaling parameters in the model along the direction of each feature vector in a Hessian matrix of feature data, the more dominant the features are, the larger the scaling is, the less secondary the features are, and the smaller the scaling is. Thus, the effect of the expanded secondary features can be reduced to a small value while the size of the primary features remains almost unchanged. The trained and verified logistic regression model is used as the final prediction model, so that the behavior of the user can be predicted, and personalized starting service is provided.
Example (b):
the method for predicting the drama chase of the user on the television based on the watching behavior in the embodiment comprises the following steps of:
a. extracting characteristics based on historical watching behavior data of a television by a user;
the feature extraction flow in this step is shown in fig. 2, and includes:
a1. time-sequencing data:
and converting the startup and shutdown time axis and the video startup and shutdown time axis into standard time stamps, sequencing the time stamps in an ascending order, and integrating startup data and video data into data according to time serialization.
a2. Data cleaning:
and effectively starting up the video once after the video is started up and watching at least 10 minutes, clearing data according to effective visual observation once after a part of video is started and watched at least 5 minutes, and clearing data according to the part unit (cover id unit) of the video.
a3. And (3) extracting all features:
and extracting all characteristics which can be calculated according to the data after data cleaning. For example, the startup duration is calculated through a startup and shutdown time axis, and the number of times that the video is watched first after being started is calculated through a video startup time axis; calculating the video watching duration according to the time axes of video starting and exiting; the number of sets of videos to be watched, the number of sections of videos to be watched every day, and the like are calculated according to the fact that cover ids of videos are the same and video ids are different.
In general, all the features that can be extracted are as follows, all features being indexed according to a video viewed by a television terminal:
| feature(s) | Feature numbering |
| (the video) is continuously started up and is watched for the first time | ① |
| Number of days of watching | ② |
| Number of "parts" of the viewed video (on the day the video was viewed) | ③ |
| Number of times (the video) was viewed | ④ |
| (the video) number of watched episode | ⑤ |
| Total length of time watched | ⑥ |
| Number of sets left (of the video) | ⑦ |
| Total number of sets (of the video) | ⑧ |
a4. Data was expected to be 0:
one tenth of the characteristic data is randomly extracted according to rows (the original data volume is large), the expected value of each column is obtained according to the column unit of the extracted characteristic data, and then the expected value is subtracted from all elements in the column. After the operation is completed, the expectation of all columns is 0, and the expectation of the entire characteristic data matrix is 0.
a5. Extracting main features through principal component analysis, and performing feature combination:
the data set expected to be 0 is subjected to a principal component analysis method, and 5-dimensional important data are left. The PCA method is characterized in that a program is completed by calling a covariance class and a feature decomposition class in a numpy library through python according to a general principle, then a data set is matrixed and input into an algorithm to obtain five-dimensional data with the maximum feature value, and the features of an object are main features to be reserved.
The reserved characteristics are ③, ④, ⑤, ⑥ and ⑧ after operation
According to the common sense judgment, the two-dimensional features ⑤ and ⑥ are linearly related in most cases of series (a small part of people watch a certain set to fast forward), namely y is ax, wherein y is watching time length, x is set number, and a is time length of each set.
Since ⑤ and ⑧ are both the number of sets, one is the number of viewing sets and one is the total number of sets, the combination of ⑤ divided by ⑧ makes up the progress of viewing the video, and this feature holds if the total number of sets is at least greater than 10 sets.
Thus, the entire feature engineering contains 3-dimensional features: viewing times, viewing progress, total number of videos viewed on the same day.
b. Making a data set by using the extracted data;
and c, making a data set for training and verification according to the main feature dimension finally determined in the step a and the extracted feature data, wherein the making process is shown in fig. 3 and comprises the following steps:
b1. traversing all data of all televisions in an observation time period, listing all main characteristics of videos with the total collection number of each television watched being more than 10 sets, and arranging the main characteristics in ascending order according to a time axis;
b2. according to the fact that the watching times are more than 3 times, the watching collection number is more than 5 collections, the video is watched when the television is started next time, the video is defined as a drama chase, a label is made, if the value of the label is 1, the video is continuously watched when the television is started next time, if the value of the label is 0, the video is not watched when the television is started next time, and the characteristic data and the label data are correspondingly associated;
b3. normalizing the characteristic data: solving the maximum value of each dimension of features, and dividing each dimension of features by the maximum value;
b4. calling a characteristic polynomial expansion module in a sklern library, and performing polynomial expansion on the characteristic data to expand the characteristic data into a multi-order and mutually associated characteristic set;
b5. and constructing an empty training data set and a verification data set, and then randomly distributing all feature data and label data into the training data set and the verification data set according to the proportion of 7: 1.
b6. And correspondingly outputting the training characteristic data, the training label data, the verification characteristic data and the verification label data to a 4-text file.
c. Inputting the manufactured data set into a logistic regression model for training and verification;
calling Logistic regression class in skleern by python language to compile a logistic regression model, selecting random average gradient descent 'sag' as an optimization method, and selecting second-order norm regularization (L2 regularization) to input data in a training data set (including training characteristic data and training label data) into the logistic regression model for training in order to avoid overfitting; and inputting the data in the verification data set (comprising verification feature data and verification tag data) into a logistic regression model for verification so as to enable the precision to reach the standard.
d. And predicting the viewing behavior of the user by using the trained and verified logistic regression model.
The logistic regression model after continuous training and verification is basically in a stable state and can be used as a prediction model to predict the watching behavior of the user, so that personalized startup service or personalized recommendation service is provided for the user.
Claims (7)
1. A method for predicting a user's drama chase on a television based on viewing behavior, comprising the steps of:
a. extracting characteristics based on historical watching behavior data of a television by a user;
b. making a data set by using the extracted data;
c. inputting the manufactured data set into a logistic regression model for training and verification;
d. predicting the watching behavior of the user by utilizing the trained and verified logistic regression model;
in step a, the historical watching behavior data of the user on the television comprises:
video playing data: the method comprises the steps of a time axis of a user entering a video player and a time axis of starting and exiting a certain video;
television on-off data: the television comprises a television startup time axis and a television shutdown time axis;
media information: the video information for playing comprises the name of the played video, the ID of a video series, the type of the played video and the brief introduction;
the feature extraction specifically includes:
a1. converting a television startup and shutdown time axis and a video startup and exit time axis into standard time stamps, sequencing the time stamps in an ascending order, and integrating the time stamps into data according to time serialization;
a2. flushing invalid data, the invalid data comprising: data with too short starting time and video playing time;
a3. extracting characteristics which can be calculated according to the cleaned data;
a4. carrying out expectation processing on the data, and processing the expected value of the data into 0;
a5. and selecting main characteristics from the data subjected to the anticipatory processing by adopting a principal component analysis method, and combining the characteristics.
2. The method of predicting a series of user's events on a television based on viewing behavior as claimed in claim 1, wherein step d further comprises: and if the situation that the user is in the episode chase is predicted, directly jumping to the latest watching progress of the episode traced by the user for video playing after the next startup.
3. The method of predicting a series of user's trails on tv based on viewing behavior as claimed in claim 1, wherein in step a3, said calculable features comprise: calculating the starting-up time length through a starting-up and closing-down time axis, and calculating the first watching times of the video after the video is started up through a video starting time axis; calculating the video watching duration according to the time axes of video starting and exiting; the number of sets of videos to be watched and the number of sections of videos to be watched per day are calculated based on the fact that the series IDs of videos are the same and the video IDs are different.
4. The method according to claim 1, wherein the step a4 of anticipating the episode of the user's tv show based on the viewing behavior, wherein the step a4 of anticipating the data to obtain an expectation value of 0 includes:
and c, randomly extracting a certain amount of feature data from the feature data extracted in the step a3 according to rows, calculating an expected value of each column according to the extracted feature data by column units, subtracting the expected values from all elements in the column, and after the operation is finished, the expectation of all the columns is 0, so that the expectation of the whole feature data matrix is 0.
5. The method according to claim 1, wherein the step a5 of selecting principal features from the data after expectation processing by principal component analysis and combining the features comprises: and performing characteristic decomposition on the characteristic data matrix to obtain a characteristic value list of the matrix, wherein each characteristic value corresponds to a one-dimensional characteristic vector, the larger the characteristic value is, the more important the dimensional characteristic vector is, otherwise, the smaller the characteristic value is, the less important the dimensional characteristic vector is, the n-dimensional characteristic with the maximum characteristic value is taken as a main characteristic, and then, characteristic combination is performed according to the linear correlation relationship among the selected main characteristics, so that the final characteristic dimension is determined.
6. The method of predicting a series of user's trails on tv based on viewing behavior as claimed in claim 5, wherein the step b of using the extracted data to produce a data set specifically comprises:
b1. traversing all data of all televisions in an observation time period, listing all main characteristics of videos with the total collection number of each television watched being more than 10 sets, and arranging the main characteristics in ascending order according to a time axis;
b2, according to the condition that the watching times are more than 3 times, the watching collection number is more than 5 collections, the video is watched at the next starting, the video is defined as a series, a label is made, if the label value is 1, the video is continuously watched at the next starting, if the label value is 0, the video is not watched at the next starting, and the characteristic data and the label data are correspondingly associated;
b3. normalizing the characteristic data: solving the maximum value of each dimension of features, and dividing each dimension of features by the maximum value;
b4. calling a characteristic polynomial expansion module in a sklern library, and performing polynomial expansion on the characteristic data to expand the characteristic data into a multi-order and mutually associated characteristic set;
b5. constructing an empty training data set and a verification data set, and then randomly distributing all feature data and label data into the training data set and the verification data set according to the proportion of 7: 1;
b6. and correspondingly outputting the training characteristic data, the training label data, the verification characteristic data and the verification label data to a 4-text file.
7. The method for predicting user chase after on tv of any one of claims 1-6, wherein in step c, the generated data set is input into a logistic regression model to be trained, and a second-order norm regularization is added into the logistic regression model.
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 CN107635143A (en) | 2018-01-26 |
| CN107635143B true 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) |
Families Citing this family (5)
| 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 |
| 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 |
| 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 |
| CN116017077A (en) * | 2022-12-12 | 2023-04-25 | 北京奇艺世纪科技有限公司 | Direct playing method and device, electronic equipment and storage medium |
| 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 (7)
| 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 |
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9473803B2 (en) * | 2014-08-08 | 2016-10-18 | TCL Research America Inc. | Personalized channel recommendation method and system |
-
2017
- 2017-11-06 CN CN201711078604.8A patent/CN107635143B/en active Active
Patent Citations (7)
| 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107635143A (en) | 2018-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107635143B (en) | Method for predicting user's drama chase on television based on watching behavior | |
| CN110781391B (en) | Information recommendation method, device, equipment and storage medium | |
| WO2019043352A1 (en) | System and method for content discovery | |
| US9473803B2 (en) | Personalized channel recommendation method and system | |
| CN104991966B (en) | Media content ordering method and system | |
| CN101281540B (en) | Apparatus, method and computer program for processing information | |
| CN107766360B (en) | Video heat prediction method and device | |
| US20150039608A1 (en) | Media content rankings for discovery of novel content | |
| US20140123179A1 (en) | Zoomable content recommendation system | |
| CN112464025B (en) | Video recommendation method and device, electronic equipment and medium | |
| KR20140135207A (en) | Automatically recommending content | |
| CN104994426A (en) | Method and system of program video recognition | |
| CN109511015B (en) | Multimedia resource recommendation method, device, storage medium and equipment | |
| CN101047831A (en) | Recommendation program information method and device | |
| CN103052953A (en) | Information processing device, method of processing information, and program | |
| US20150212922A1 (en) | Data analysis apparatus, method and program | |
| US20230066853A1 (en) | Method and apparatus for training information prediction models, method and apparatus for predicting information, and storage medium and device thereof | |
| CN105139021B (en) | A kind of method and system for realizing TV user Fast Classification based on rough set theory | |
| US11871060B2 (en) | Systems and methods for media content navigation and filtering | |
| CN112507163A (en) | Duration prediction model training method, recommendation method, device, equipment and medium | |
| CN104899331A (en) | Television used behavior data clustering method and device and Spark big data platform | |
| CN112464100A (en) | Information recommendation model training method, information recommendation method, device and equipment | |
| US11252375B2 (en) | Electronic apparatus and control method thereof | |
| CN112199582A (en) | Content recommendation method, device, equipment and medium | |
| US20190286744A1 (en) | Zoomable user interface for tv |
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 |