Unified attractiveness prediction framework based on content impact factor

• US 9,875,443 B2
• Filed: 06/18/2015
• Issued: 01/23/2018
• Est. Priority Date: 06/18/2015
• Status: Active Grant

• Current Assignee: TCL Research America, Inc. (TCL Technology Group Corp.)
• Original Assignee: TCL Research America, Inc. (TCL Technology Group Corp.)
• Inventors: Ding, Wanying, Guo, Lifan, Shang, Yue, Wang, Haohong
• Primary Examiner(s): Holmes, Michael B
• Application Number:
  US14/743,141
• Publication Number:
  US 20160371598A1
• Time in Patent Office: 950 Days
• Field of Search: 706 52
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G11B 27/00 Editing; Indexing; Addressing; Timing or synchronising; Monitoring; Measuring tape travel
• CPC Class Codes: H04N 21/2407 Monitoring of transmitted content, e.g. distribution time, number of downloads arrangements for monitoring programmes for broadcast or for distribution combined with broadcast H04H20/14
• CPC Class Codes: H04N 21/251 Learning process for intelligent management, e.g. learning user preferences for recommending movies details of learning user preferences for the retrieval of video data in a video database G06F16/739; computer systems using learning methods G06N3/08
• CPC Class Codes: H04N 21/4788 communicating with other users, e.g. chatting arrangements for providing for computer conferences, e.g. chat rooms, to substation in data switching networks H04L12/1813; distributed application using peer-to-peer [P2P] networks H04L67/104
• CPC Class Codes: H04N 21/4826 using recommendation lists, e.g. of programs or channels sorted out according to their score
• CPC Class Codes: H04N 21/80 Generation or processing of content or additional data by content creator independently of the distribution process; Content per se arrangements for generating broadcast information H04H60/02

Abstract

A unified attractiveness prediction method is provided. The method includes receiving a plurality of videos and extracting at least one of metadata and view data from each of the plurality of received videos, wherein the metadata is information for describing video contents, and the view data is a total number of users who watch the video. The method also includes obtaining potential view amounts of the plurality of received videos and calculating impact factor scores of the plurality of received videos based on the potential view amounts, each impact factor score including a numerical score for indicating a degree of effectiveness of a corresponding video. Further, the method includes providing a video with a highest impact factor score based on the calculated impact factor scores.

Claims

1. A unified attractiveness prediction method, comprising:

receiving a plurality of videos;

extracting at least one of metadata and view data from each of the plurality of received videos, wherein the metadata is information for describing video contents, and the view data is a total number of users who watch the video;

based on the extracted view data, performing a Hawkes process for the received videos associated with the view data to infer a plurality of parameters of the received videos;

based on the plurality of obtained parameters and features converted from the extracted metadata, creating a linear regression model and learning feature weight parameters from the created linear regression model;

obtaining potential view amounts of the plurality of received videos;

based on the potential view amounts, calculating impact factor scores of the plurality of received videos, each impact factor score including a numerical score for indicating a degree of effectiveness of a corresponding video; and

based on the calculated impact factor scores, providing a video with a highest impact factor score.

2. The method according to claim 1, wherein:

the plurality of obtained parameters include immigrant parameters and propagation parameters, wherein the immigrant parameters govern a total number of users who spontaneously watch the video, and the propagation parameters govern a total number of users who watch the video by others'"'"' recommendation.

3. The method according to claim 1, wherein creating a linear regression model of the received videos further includes:

using the obtained immigrant parameters as dependent variables of the linear regression model; and

using the features converted from the metadata as independent variables of the linear regression model.

4. The method according to claim 1, wherein:

a plurality of nodes, including positive nodes and negative nodes, corresponding to the plurality of received videos compose the Hawkes processes, wherein each of the positive nodes composes a positive Hawkes process, and each of the negative nodes composes a negative Hawkes process.

5. The method according to claim 2, further including:

when the videos include the metadata and the view data, performing the Hawkes process by using directly the view data to infer the immigrant parameters and the propagation parameters;

when the videos only include the metadata, generating the immigrant parameters by using directly the feature weight parameters for the Hawkes processes, and ignoring the propagation part, such that the immigrant parameters are used to make a rough prediction for near future; and

when the videos only include the view data, performing the Hawkes process directly and ignoring the immigrant part to make the rough prediction.

6. The method according to claim 5, wherein:

a rate function of the Hawkes process is defined by;

λ

_i=μ

_i(t)+Σ

<sub>ti,k<

t</sub>φ

_ii(t−

t_ik)+Σ

<sub>tjl<

t</sub>φ

_ij(t−

t_jl)

wherein i,jε

(1,2), and 1 indicates that the node comes from a positive queue, and 2 indicates that the node comes from a negative queue;

t is a time point;

μ

_i(t) is an immigrant function;

Σ

<sub>ti,k<

t</sub>φ

_ii(t−

t_ik) is a self-exciting function that measures how a previous event in the queue evokes other events happening in a same queue; and

Σ

<sub>tjl<

t</sub>φ

_ij(t−

t_jl) is a cross-exciting function that measures how events in one queue affect events in the other queue.

7. The method according to claim 1, wherein:

provided that the most recent K time periods data is taken into consideration, a content impact factor score (CIFS) is calculated by;

8. A system for making a unified attractiveness prediction, comprising:

a processor;

a memory coupled to the processor;

a plurality of program modules stored in the memory to be executed by the processor, the plurality of program modules comprising;

a bivariate Hawkes processing module configured to;

extract view data from a plurality of received videos, wherein the view data is a total number of users who watch the video; and

based on the extracted view data, perform a Hawkes process for the received videos associated with the view data to infer a plurality of parameters of the received videos;

a potential view amount calculation module configured to obtain potential view amounts of the plurality of received videos; and

an impact factor score calculation module configured to calculate impact factor scores of the plurality of received videos based on the potential view amounts and provide a video with a highest impact factor score based on the calculated impact factor score of each video, wherein each impact factor score includes a numerical score for indicating a degree of effectiveness of a corresponding video.

9. The system according to claim 8, further including:

a parameter inference module configured to receive the plurality of parameters inferred from the Hawkes process;

a liner regression module configured to create a linear regression model based on the plurality of obtained parameters and features converted from metadata of the received videos, wherein the metadata is information for describing video contents; and

a feature weight parameter module configured to learn feature weight parameters from the created linear regression model.

10. The system according to claim 9, wherein:

the plurality of obtained parameters include immigrant parameters and propagation parameters, wherein the immigrant parameters govern a total number of users who spontaneously watch the video, and the propagation parameters govern a total number of users who watch the video by others'"'"' recommendation.

11. The system according to claim 9, wherein the liner regression module is further configured to:

use the obtained immigrant parameters as dependent variables of the linear regression model; and

use the features converted from the metadata extracted from the plurality of received videos as independent variables of the linear regression model.

12. The system according to claim 8, wherein:

a plurality of nodes, including positive nodes and negative nodes, corresponding to the plurality of received videos compose the Hawkes processes, wherein each of the positive nodes composes a positive Hawkes process, and each of the negative nodes composes a negative Hawkes process.

13. The system according to claim 9, wherein:

when the videos include the metadata and the view data, the Hawkes process is performed by using directly the view data to infer the immigrant parameters and the propagation parameters;

when the videos only include the metadata, the immigrant parameters is generated by using directly the feature weight parameters for the Hawkes processes, and the propagation part is ignored, such that the immigrant parameters are used to make a rough prediction for near future; and

when the videos only include the view data, the Hawkes process is performed directly, and the immigrant part is ignored, making the rough prediction.

14. The system according to claim 13, wherein:

a rate function of the Hawkes process is defined by;

λ

_i=μ

_i(t)+Σ

<sub>ti,k<

t</sub>φ

_ii(t−

t_ik)+Σ

<sub>tjl<

t</sub>φ

_ij(t−

t_jl)

wherein i,jε

(1,2), and 1 indicates that this node comes from a positive queue, and 2 indicates that this node comes from a negative queue;

t is a time point;

μ

_i(t) is an immigrant function;

Σ

<sub>ti,k<

t</sub>φ

_ii(t−

t_ik) is a self-exciting function that measures how a previous event in the queue evokes other events happening in a same queue; and

Σ

<sub>tjl<

t</sub>φ

_ij(t−

t_jl) is a cross-exciting function that measures how events in one queue affect events in the other queue.

15. The system according to claim 8, wherein:

provided that the most recent K time periods data is taken into consideration, a content impact factor score (CIFS) is calculated by;