WO2019109592A1

WO2019109592A1 - Smart video recommendation method and system

Info

Publication number: WO2019109592A1
Application number: PCT/CN2018/086407
Authority: WO
Inventors: 傅金澍
Original assignee: 上海斐讯数据通信技术有限公司
Priority date: 2017-12-07
Filing date: 2018-05-10
Publication date: 2019-06-13
Also published as: CN108134950B; CN108134950A

Abstract

Disclosed in the present invention are a smart video recommendation method and system, for use in resolving the problem in the prior art that related video recommendation is given without determining whether a user likes a watched video. The smart video recommendation method comprises the steps of: S1, determining whether playing time of multiple videos falls within a preset time range, according to video playing records of users; S2, marking the multiple videos by using corresponding preset labels; S3, establishing a preset video recommendation model; and S4, computing a popularity among users corresponding to each video in a video library according to the established preset video recommendation model, arranging the videos in the video library according to the popularities, and displaying videos ranking preset top preference positions in a video recommendation bar. In the video recommendation method, a more intelligent and personalized video recommendation model can be established according to watching records of users, so that videos preferred by the users are recommended, thereby improving the experience effect for watching videos online by the users.

Description

Intelligent video recommendation method and system

The present application claims the priority of the Chinese patent application filed on Dec. 7, 2017, which is hereby incorporated by reference.

Technical field

The present invention relates to the field of intelligent video recommendation technologies, and in particular, to an intelligent video recommendation method and system.

Background technique

When the user watches the video online, they will see the video recommendation column, but the user will find that the recommended video is not what you like, and most of them are not interested. The system does not judge well that the user likes to watch the video and gives the relevant video recommendation, which affects the user's experience. At the same time, the system recommends a lot of recommended videos only with the video that the user has watched. Similar, they do not take into account the characteristics of the video, and thus do not achieve the best experience.

For example, Chinese Patent Publication No. 102843586B discloses a video recommendation method, including: acquiring a target recommendation video determined according to a user operation record; setting a video recommendation list having the target recommendation video; and transmitting the video recommendation to the user List. The invention provides a video recommendation method for determining a target recommendation video according to an operation record of a single user, the target recommendation video including an unwatched video added by the user to a favorite or a bookmark, or the user click frequency satisfying the preset click frequency a certain category of video; since the target recommendation video is determined according to a personal operation record of a single user, and the personal operation record reflects the personal preference of the individual user to a certain extent; therefore, when the target recommendation video is sent to the user It enables users to more directly access their favorite videos.

However, the video recommended by the video recommendation system proposed in the prior art is not necessarily the user's favorite. The system does not judge the user's favorite video or not, and gives the relevant video recommendation, which affects the user experience. At the same time, many recommended videos given by the system are only similar to the videos that the user has watched, and do not comprehensively consider the characteristics of the video, so that the best experience is not achieved.

Therefore, in summary, an intelligent video recommendation method and system need to be proposed, which can systematically analyze and determine the recommendation of the video.

Summary of the invention

The object of the present invention is to provide an intelligent video recommendation method and system for the defects in the prior art. The intelligent video recommendation method uses a dichotomy algorithm to synthesize various features of the video, establishes a video recommendation model, and recommends according to the established video. The model and the video library are recommended to the video that the user likes, and the recommendation for accurately judging the user's favorite video is realized, and the user's viewing video experience is improved.

In order to achieve the above object, the present invention adopts the following technical solution:

An intelligent video recommendation method includes the steps of:

S1: determining, according to the video play record of the user, whether a play time of the multiple videos in the video play record is within a preset time range;

S2: marking a video whose playing time exceeds a preset time range by a first preset label, and marking a video whose playing time is within a preset time range by a second preset label;

S3: establishing a preset video recommendation model according to preset features of each of the multiple videos and preset tags corresponding to the tags;

S4: In the video library, according to the established preset video recommendation model, calculate the corresponding user likeness of each video, arrange the videos in the video library according to the user's likeness, and display the arrangement in the video recommendation column. The video with the preset number of digits.

Further, the step S1 includes:

S11: Obtain a video play record of the user;

S12: Record a play time corresponding to each video according to the obtained user video play record;

S13: Determine whether the play time corresponding to each video is within a preset time range.

Further, the step S2 includes:

S21: classify the multiple videos, divide the video whose playing time exceeds the preset time range into videos that the user likes, and divide the videos whose playing time is within the preset time range into videos that the user does not like;

S22: Mark the video of the above category, mark the video that the user likes as the first preset label, and mark the video that the user does not like as the second preset label.

Further, the established preset video recommendation model is:

Further, the step S4 includes:

S41: Collect preset features of each video in the video library;

S42: Calculate a user preference degree corresponding to each video according to the preset feature of each captured video and the preset video recommendation model that is established;

S43: arranging the videos in the video library according to a user's preference according to a preset arrangement condition;

S44: Display the video of the preset number of digits in the video recommendation column.

An intelligent video recommendation system comprising:

a judging time module, configured to determine, according to a video play record of the user, whether a play time of the plurality of videos in the video play record is within a preset time range;

a tag labeling module, configured to mark a video whose playing time exceeds a preset time range by a first preset label, and mark a video whose playing time is within a preset time range by a second preset label;

Establishing a model module, configured to establish a preset video recommendation model according to a preset feature of each of the plurality of videos and a preset tag of the corresponding tag;

The recommended video module is configured to calculate a corresponding user likeness of each video according to the established preset video recommendation model in the video library, arrange the videos in the video library according to the user's likeness, and recommend in the video. The column shows the video of the preset number of digits.

Further, the determining time module includes:

An obtaining unit, configured to acquire a video play record of the user;

a recording unit, configured to record a play time corresponding to each video according to the acquired user video play record;

The determining unit is configured to determine whether the playing time corresponding to each video is within a preset time range.

Further, the tag label module includes:

a classifying unit, configured to classify the plurality of videos into videos that are played by the user in a preset time range, and divide the videos whose playing time is within a preset time range into videos that the user does not like;

a marking unit, configured to mark the video of the above category, mark the video that the user likes as the first preset label, and mark the video that the user does not like as the second preset label.

Further, the building model module includes:

a statistical unit, configured to count preset features of each video and preset labels of corresponding tags;

And a calculating unit, configured to calculate a preset video recommendation model according to preset features of each video.

Further, the recommended video module includes:

An acquisition unit, configured to collect preset features of each video in the video library;

Calculating a user likeness unit, configured to calculate a user likeness corresponding to each video according to the preset feature of each captured video and the established preset video recommendation model;

An arranging unit, configured to arrange the videos in the video library according to a user's preference according to a preset arrangement condition;

A recommended display unit for displaying the video of the preset preset number of bits in the video recommendation field.

DRAWINGS

1 is a flow chart 1 of an intelligent video recommendation method according to the present invention;

2 is a second flowchart of a smart video recommendation method according to the present invention;

3 is a structural diagram 1 of the intelligent video recommendation system of the present invention;

4 is a structural diagram 2 of the intelligent video recommendation system of the present invention.

Detailed ways

The technical solutions of the present invention are further described below with reference to the accompanying drawings, but the present invention is not limited to the embodiments.

Embodiment 1

This embodiment provides an intelligent video recommendation method. As shown in FIG. 1 , the method includes the following steps:

The main problem solved by the intelligent video recommendation method provided by this embodiment is that when we watch the video, we are willing to watch the video that we like, but for the video that we don't like, we will not look at it for a shorter time.

The smart video recommendation method provided in this embodiment divides the watched videos into two categories according to the user's viewing record, one is a favorite video, the other is a video that is not like, and the favorite video is marked as the first pre- Marking, the video that you don't like is marked as the second preset mark, which introduces the concept of two-category; after the mark, the corresponding feature is used to train the video recommendation model; then the trained video recommendation model is used in the video library. The video is used to estimate the user's likeness; then the video is sorted according to the user's likeness, the user likes the high ranks in the front, and the user likes the low ranks in the back; finally, the top ranked videos are recommended.

The specific method is:

Firstly, according to the video play record of the user, it is determined whether the playing time of the plurality of videos is within a preset time range; for example, in the video recording, the user watches the video for ten minutes, the user watches the video for one minute, and the user watches the video. After watching for 50 seconds, for example, the preset time range is 2 minutes, the play time of the video one exceeds the preset time range, and the play time of the video 2 and the video 3 is within the preset time range.

And then the video of the plurality of videos whose playing time exceeds the preset time range is marked with the first preset label, and the video of the plurality of videos whose playing time is within the preset time range is marked with the second preset label, for example, The above video marks the first preset label, and the video 2 and the video 3 mark the second preset label.

Establishing a preset video recommendation model according to a preset feature of each of the plurality of videos and a preset tag of the corresponding mark, for example, a preset feature of the video one: the video type is a motivational film, the director is Feng Xiaogang, and the actor is Zhang Yixing. In the 1990s, and the first preset label for the video markup, a preset video recommendation model is established, that is, according to the preset features of the video 1, the video 2, the video 3, and all the videos in the viewing record, and the preset label. Calculate the video recommendation formula.

In the video library, the preset features of each video are substituted into the established preset video recommendation model, and the corresponding user likeness of each video in the video library is calculated, and the videos in the video library are arranged according to the likeness. And display the video of the preset number of presets in the video recommendation bar.

The intelligent video recommendation method provided in this embodiment introduces the two-category concept into the video recommendation system, and solves the user video recommendation problem better and more easily.

Embodiment 2

This embodiment provides an intelligent video recommendation method. As shown in FIG. 2, the method provided in this embodiment adds the following steps to the first embodiment:

Further, the step S1 includes:

S11: Obtain a video play record of the user;

Further, the step S2 includes:

Further, the established preset video recommendation model is:

Further, the step S4 includes:

S41: Collect preset features of each video in the video library;

The specific implementation steps are as follows:

Firstly, the user's video play record is obtained, and according to the obtained user video play record, the play time corresponding to each video is recorded, and multiple videos in the video play record are respectively judged to determine whether the play time is within a preset time range;

And then classifying the plurality of videos into videos that are played by the user in a predetermined time range, and dividing the videos whose playing time is within the preset time range into videos that the user does not like;

The video of the above category is marked, the video that the user likes is marked as the first preset label, and the video that the user does not like is marked as the second preset label.

A preset video recommendation model is established according to a preset feature of each of the plurality of videos and a preset tag of the corresponding tag.

Specifically:

A video has the following characteristics:

(1) Type: plot, martial arts, spy war, city, costume, family, funny, love, etc.;

(2) Regions: Mainland, Hong Kong, Taiwan, Europe, America, Korea, Japan, etc.;

(3) Year: 2017, 2016, 2015, 2014, 2013, 2012, 2011, etc.;

(4) Directors: Zhang Yimou, Guan Hu, Chen Kaige, Feng Xiaogang, Xu Ke, Jiang Wen, etc.;

(5) Actors: Liu Shishi, Huo Jianhua, Zhao Liying, Sun Wei, Chen Xiao, and Dai Dawei.

In order to better represent the characteristics of the video, the present invention expresses the above characteristics: type, region, age, director, main actors, etc. as x ₁ , x ₂ , x ₃ , ..., x _n , where n represents the total of the features The number of each feature is as described above. For the video in the user's viewing record, when the viewing time exceeds S minutes, the user is considered to like the video, and the relevant features of the video can be extracted for training the model, and the corresponding label of the video is set to 1 When the viewing time does not exceed S minutes, the user is considered to dislike the video, and the relevant features of the video can be extracted for training the model, and the label corresponding to the video is set to zero. The video recommendation model is expressed as:

among them,

z _j =w ₁ *x ₁ +w ₂ *x ₂ +...+w _n *x _n +b (1-2)

Where w ₁ , w ₂ , w ₃ , . . . , w _n represent the weight coefficient corresponding to each feature, and b represents the offset coefficient, which is mainly used to limit the size of the weight coefficient, and is estimated in the user preference degree. In the stage, j represents the jth video in the video library, and Z represents the weighting value of each feature of the video j. Assuming that the number of videos of the entire video library is M ₁ , z _j represents the weighted value of each feature of the video j. Assuming that the number of videos in the entire video library is M ₁ , then j ∈ [1, M ₁ ], y _j is the user's likeness, indicating the degree of user's preference for the video z _j , and its value range is (0, 1) .

The algorithm steps of the system are:

First, the model (1-1) is trained by the user's play record. When the user plays the video j for more than S, the user is considered to like the video, and the corresponding label of the video is set to 1, that is, y _j =1, when the time when the user plays the video j does not exceed S, the user is considered to dislike the video, and the corresponding label of the video is set to 0, that is, y _j =0, and the whole training process utilizes the maximum Likelihood estimation method

Then, using the trained model, the user's likeness estimation is performed on the video in the video library. The closer the estimated value is to 1, the higher the user's preference is, and the closer the estimated value is to 0, the lower the user's preference.

Then, according to the level of user preference, the videos are sorted, the user likes the high ranks in the front, the user likes the low ranks behind, and gives the top N videos, and the corresponding videos are given in the recommendation column.

The present embodiment provides an intelligent video recommendation method, which can recommend a video that the user likes to watch according to the user's viewing record, and solves the problem that the video recommendation type is single when the user views the video. This embodiment can be based on the comprehensive feature of the user watching the video. To recommend a video that users prefer, when considering a video, consider more comprehensive.

Embodiment 3

This embodiment provides an intelligent video recommendation system. As shown in FIG. 3 and FIG. 4, the system provided in this embodiment includes:

Further, the determining time module includes:

An obtaining unit, configured to acquire a video play record of the user;

Further, the tag label module includes:

a classifying unit, configured to classify the plurality of videos into videos of a user's favorite video, and divide the videos whose playing time is within a preset time range into users that are not liked by the user. Video category;

Further, the building model module includes:

Further, the recommended video module includes:

Firstly, according to the obtaining unit in the determining time module, the video playing record of the user is obtained; then the recording unit plays the record according to the obtained user video, and records the playing time corresponding to each video; the determining unit plays according to each video recorded. The time determines whether the playback time corresponding to each video is within a preset time range.

Then, the plurality of videos are classified by the classification unit of the tag module, and the videos in which the playback time exceeds the preset time range are divided into video categories that the user likes, and the multiple videos are included. The video whose playback time is within the preset time range is classified into a video that the user does not like; and the marking unit is used to mark the video of the above category, and mark the video of the type that the user likes as the first pre- A label is set to mark the video that the user does not like as a second preset label.

And establishing a model module, wherein the establishing model module calculates a preset feature of each video and a preset label of the corresponding mark by using a statistical unit; and then calculating, by the calculating unit, the preset video recommendation model according to the preset feature of each video.

Finally, the recommended video module is obtained, and the preset feature of each video in the video library is collected by the collecting unit of the recommended video module; and then the preset feature of each video collected and the preset video recommendation are calculated by calculating the user favorite unit. The model calculates the user's likeness corresponding to each video; and arranges the videos in the video library according to the user's preference according to the preset arrangement order by the arranging unit; and presets the preset number of bits in the preset order by the recommended display unit The video is displayed in the video recommendation bar.

The intelligent video recommendation system provided by the embodiment can distinguish the video that the user likes or dislikes according to the user's viewing record, and trains more intelligent and personalized video recommendation according to the characteristics corresponding to the video that is liked and disliked. The model, which recommends videos that users prefer, thus improving the user experience when watching videos online.

The specific embodiments described herein are merely illustrative of the spirit of the invention. A person skilled in the art can make various modifications or additions to the specific embodiments described or in a similar manner, without departing from the spirit of the invention or as defined by the appended claims. The scope.

Claims

An intelligent video recommendation method, comprising the steps of:

S1: determining, according to the video play record of the user, whether a play time of the multiple videos in the video play record is within a preset time range;

S2: marking a video whose playing time exceeds a preset time range by a first preset label, and marking a video whose playing time is within a preset time range by a second preset label;

S3: establishing a preset video recommendation model according to preset features of each of the multiple videos and preset tags corresponding to the tags;

S4: In the video library, according to the established preset video recommendation model, calculate the corresponding user likeness of each video, arrange the videos in the video library according to the user's likeness, and display the arrangement in the video recommendation column. The video with the preset number of digits.
The intelligent video recommendation method according to claim 1, wherein the step S1 comprises:

S11: Obtain a video play record of the user;

S12: Record a play time corresponding to each video according to the obtained user video play record;

S13: Determine whether the play time corresponding to each video is within a preset time range.
The intelligent video recommendation method according to claim 1, wherein the step S2 comprises:

S21: classify the multiple videos, divide the video whose playing time exceeds the preset time range into videos that the user likes, and divide the videos whose playing time is within the preset time range into videos that the user does not like;

S22: Mark the video of the above category, mark the video that the user likes as the first preset label, and mark the video that the user does not like as the second preset label.
The intelligent video recommendation method according to claim 1, wherein the established preset video recommendation model is:
The intelligent video recommendation method according to claim 1, wherein the step S4 comprises:

S41: Collect preset features of each video in the video library;

S42: Calculate a user preference degree corresponding to each video according to the preset feature of each captured video and the preset video recommendation model that is established;

S43: arranging the videos in the video library according to a user's preference according to a preset arrangement condition;

S44: Display the video of the preset number of digits in the video recommendation column.
An intelligent video recommendation system, comprising:

a judging time module, configured to determine, according to a video play record of the user, whether a play time of the plurality of videos in the video play record is within a preset time range;

a tag labeling module, configured to mark a video whose playing time exceeds a preset time range by a first preset label, and mark a video whose playing time is within a preset time range by a second preset label;

Establishing a model module, configured to establish a preset video recommendation model according to a preset feature of each of the plurality of videos and a preset tag of the corresponding tag;

The recommended video module is configured to calculate a corresponding user likeness of each video according to the established preset video recommendation model in the video library, arrange the videos in the video library according to the user's likeness, and recommend in the video. The column shows the video of the preset number of digits.
The intelligent video recommendation system according to claim 6, wherein the determining time module comprises:

An obtaining unit, configured to acquire a video play record of the user;

a recording unit, configured to record a play time corresponding to each video according to the acquired user video play record;

The determining unit is configured to determine whether the playing time corresponding to each video is within a preset time range.
The intelligent video recommendation system according to claim 6, wherein the tag label module comprises:

a classifying unit, configured to classify the plurality of videos into videos that are played by the user in a preset time range, and divide the videos whose playing time is within a preset time range into videos that the user does not like;

a marking unit, configured to mark the video of the above category, mark the video that the user likes as the first preset label, and mark the video that the user does not like as the second preset label.
The intelligent video recommendation system according to claim 6, wherein the building model module comprises:

a statistical unit, configured to count preset features of each video and preset labels of corresponding tags;

And a calculating unit, configured to calculate a preset video recommendation model according to preset features of each video.
The intelligent video recommendation system according to claim 6, wherein the recommended video module comprises:

An acquisition unit, configured to collect preset features of each video in the video library;

Calculating a user likeness unit, configured to calculate a user likeness corresponding to each video according to the preset feature of each captured video and the established preset video recommendation model;

An arranging unit, configured to arrange the videos in the video library according to a user's preference according to a preset arrangement condition;

A recommended display unit for displaying the video of the preset preset number of bits in the video recommendation field.