CN112989115B

CN112989115B - Screening control method and device for video to be recommended

Info

Publication number: CN112989115B
Application number: CN202110153744.7A
Authority: CN
Inventors: 陈畅新; 钟艺豪; 李百川
Original assignee: Youmi Technology Co ltd
Current assignee: Youmi Technology Co ltd
Priority date: 2021-02-04
Filing date: 2021-02-04
Publication date: 2024-06-25
Anticipated expiration: 2041-02-04
Also published as: CN112989115A

Abstract

The invention discloses a screening control method and device for videos to be recommended, comprising the following steps: extracting target characteristic information of a target video and determining a target class label of the target video; screening a target multi-category set comprising target category labels from a plurality of multi-category sets, and screening at least one video which meets a preset matching condition with the matching degree of target characteristic information from a video set corresponding to the target multi-category set to obtain a first video set; according to the hash characteristics of the videos, filtering all videos included in the first video set to obtain a second video set, wherein the filtering operation is used for filtering videos with the content overlapping ratio of the video pictures higher than an overlapping ratio threshold value from the first video set; and determining the videos included in the second video set as videos to be recommended of the target videos. Therefore, the method and the device can automatically filter the videos with different picture layouts and higher content overlapping ratio, improve the effectiveness of the screened video to be recommended, and are beneficial to improving the matching degree of the video to be recommended and the user requirement.

Description

Screening control method and device for video to be recommended

Technical Field

The invention relates to the technical field of video processing, in particular to a method and a device for screening and controlling videos to be recommended.

Background

With the rapid development of electronic information, short videos are gradually exploded and the corresponding user scale is also gradually expanded, so that users can directly shoot videos and clip and release the videos, different creation elements (also called artistic effects) can be added based on the design of the existing videos to re-create the existing videos, for example, the users can modify pictures, luminosity, codes and the like of the existing videos, daily lives of the users can be enriched, and certain benefits can be brought to the users in some cases. As more and more users participate in publishing videos or re-authoring existing videos, there is a large number of duplicate or related videos in a massive video database, and in order to increase user viscosity, related videos need to be recommended to users according to videos watched by the users.

Practice finds that the related video recommendation methods currently existing can be roughly divided into three types: the first is a recommendation method based on collaborative filtering, specifically, recommendation of related videos is performed according to historical viewing records of users or viewing records of other users of the same type, and the method has the problem of cold start, namely, videos which are not watched by the users cannot be recommended; the second is a recommendation method based on video content, specifically, extracting feature information of video frames and comparing associated features in a database according to feature information of target video, so as to recommend related video, wherein the method avoids the problem of cold start, but can recommend video with completely consistent content; the third method is to calculate the correlation between videos according to the label information of the videos, but the semantic features of the label information of the videos are too wide, and the video content is not directly considered, so that the recommended correlation videos cannot meet the requirements of users.

Therefore, the current video recommendation mode has the problem that the matching degree between the screened video to be recommended and the user requirement is low.

Disclosure of Invention

The invention provides a screening control method and device for videos to be recommended, which can improve the matching degree of the screened videos to be recommended and the actual demands of users.

The first aspect of the invention discloses a screening control method for videos to be recommended, which comprises the following steps:

extracting target feature information of a target video, and determining a target category label of the target video, wherein the target feature information is used for representing style features of the target video;

Screening a target multi-category set comprising the target category label from a plurality of pre-generated multi-category sets, and screening at least one video which meets a preset matching condition with the matching degree of the target characteristic information from a video set corresponding to the target multi-category set to obtain a first video set;

According to the determined hash characteristics of the target video and the hash characteristics of each video in the first video set, performing filtering operation on all videos included in the first video set to obtain a second video set, wherein the filtering operation is used for filtering out videos with the content overlapping ratio of video pictures higher than a preset overlapping ratio threshold value from all videos included in the first video set;

And determining the videos included in the second video set as videos to be recommended of the target video.

As an optional implementation manner, in the first aspect of the present invention, before the selecting, from a plurality of pre-generated multi-category sets, a target multi-category set including the target category label, the method further includes:

Storing the video identification, the feature information, the hash feature and the category label of each video in the original video set to be recommended as an information set corresponding to the video;

According to the pre-determined category label association relation, performing centralized storage operation on information sets to which all category labels with the association relation belong to, so as to obtain a plurality of multi-category sets;

Each multi-category set comprises a plurality of category labels with association relations and contents in information sets corresponding to corresponding videos.

In an optional implementation manner, in the first aspect of the present invention, according to a predetermined association relationship between category labels, a centralized storage operation is performed on information sets to which all category labels having the association relationship belong, and before obtaining a plurality of multi-category sets, the method further includes:

Generating a category label association relation;

wherein the generating the category label association relation includes:

Inputting each verification video in a verification video set into a pre-trained class identification model to obtain a plurality of class labels corresponding to each verification video, wherein each verification video has an original class label corresponding to the verification video;

Extracting category labels meeting a first preset screening condition from all category labels corresponding to each verification video to obtain a category label set corresponding to each verification video;

Determining all verification videos with the same original category labels in all the verification videos as one verification video group to obtain a plurality of verification video groups;

Screening all category labels meeting a second preset screening condition in all category labels included in the category label set corresponding to all verification videos in each verification video group to obtain a category label set corresponding to each verification video group;

and respectively establishing a category label association relationship among a plurality of category labels included in the category label set corresponding to each verification video group.

In an optional implementation manner, in a first aspect of the present invention, the screening all category labels that meet a second preset screening condition in all category labels included in the category label set corresponding to all verification videos in each verification video group, to obtain a category label set corresponding to each verification video group includes:

And counting the label number of each class label in the class label set corresponding to all the verification videos in the verification video set for each verification video set, and screening all class labels with the label number greater than or equal to a preset label number threshold value from all class labels included in the class label set corresponding to all the verification videos in the verification video set to obtain the class label set corresponding to the verification video set.

In a first aspect of the present invention, according to the determined hash feature of the target video and the hash feature of each video in the first video set, performing a filtering operation on all videos included in the first video set to obtain a second video set, where the filtering operation includes:

according to the hash characteristics of the target video and the hash characteristics of each video in the first video set, performing a first filtering operation on all videos included in the first video set to obtain a filtering result;

And executing a second filtering operation on all videos included in the filtering result according to the hash characteristics of all videos included in the filtering result to obtain a second video set.

In a first aspect of the present invention, according to the hash feature of the target video and the hash feature of each video in the first video set, a first filtering operation is performed on all videos included in the first video set to obtain a filtering result, where the filtering operation includes:

according to the hash characteristics of the target video and the hash characteristics of each video in the first video set, calculating the Hamming distance between each video in the first video set and the target video;

And filtering all videos with the Hamming distance smaller than or equal to a preset Hamming distance threshold value from the first video set to obtain a filtering result.

In an optional implementation manner, in a first aspect of the present invention, according to the hash characteristics of all videos included in the filtering result, performing a second filtering operation on all videos included in the filtering result to obtain a second video set, where the second video set includes:

selecting one of the non-selected videos from all the residual videos corresponding to the filtering result, and calculating the Hamming distance between each video except one of the videos and one of the residual videos corresponding to the filtering result according to the Hash characteristics of the one of the videos and the Hash characteristics of each video except the one of the videos in all the residual videos corresponding to the filtering result;

Judging whether all videos except one of the videos in all the remaining videos corresponding to the filtering result have at least one video with the hamming distance smaller than or equal to the preset hamming distance threshold value;

When the judgment result is negative, triggering and executing the operation of selecting one of the non-selected videos from all the residual videos corresponding to the filtering result, calculating the hamming distance between each video except one of the videos and one of the videos in all the residual videos corresponding to the filtering result according to the hash characteristics of the one of the videos and the hash characteristics of each video except the one of the videos in all the residual videos corresponding to the filtering result, and triggering and executing the operation of judging whether at least one video with the hamming distance smaller than or equal to the preset hamming distance threshold exists in all the videos except the one of the residual videos corresponding to the filtering result;

When the judgment result is yes, filtering all videos of which the Hamming distance is smaller than or equal to the preset Hamming distance threshold value from all the remaining videos of which the filtering result corresponds except for one of the videos, so as to update all the remaining videos of which the filtering result corresponds, triggering and executing the operation of selecting one of the non-selected videos from all the remaining videos of which the filtering result corresponds, according to the Hash characteristic of the one of the videos and the Hamming characteristic of each video of all the remaining videos of which the filtering result corresponds except for the one of the videos, calculating the Hamming distance between each video of all the remaining videos of which the filtering result corresponds except for the one of the videos and the one of the videos, and triggering and executing the operation of judging whether the Hamming distance between each video of all the remaining videos of which the filtering result corresponds except for the one of the videos is smaller than or equal to the preset Hamming distance threshold value;

And when only one unselected video exists in all the residual videos corresponding to the filtering result, determining all the residual videos corresponding to the filtering result as a second video set.

The second aspect of the present invention discloses a screening control device for video to be recommended, the device comprising:

The extraction module is used for extracting target feature information of a target video, wherein the target feature information is used for representing style features of the target video;

The determining module is used for determining a target category label of the target video;

The first screening module is used for screening a target multi-category set comprising the target category label from a plurality of pre-generated multi-category sets;

the second screening module is used for screening at least one video which meets the preset matching condition with the matching degree of the target characteristic information from the video set corresponding to the target multi-category set to obtain a first video set;

The filtering module is used for performing filtering operation on all videos included in the first video set to obtain a second video set according to the determined hash characteristics of the target video and the hash characteristics of each video in the first video set, wherein the filtering operation is used for filtering out videos with the content overlapping ratio of video pictures higher than a preset overlapping ratio threshold value from all videos included in the first video set;

The determining module is further configured to determine a video included in the second video set as a video to be recommended of the target video.

As an alternative embodiment, in the second aspect of the present invention, the apparatus further includes:

The multi-category set generation module is used for storing the video identification, the characteristic information, the hash characteristic and the category label of each video in the original video set to be recommended as an information set corresponding to the video before the first screening module screens the target multi-category set comprising the target category label from a plurality of pre-generated multi-category sets; according to the pre-determined category label association relation, performing centralized storage operation on information sets of all category labels with association relation to obtain a plurality of multi-category sets;

The association relation generation module is used for generating a category label association relation;

wherein, the association relation generating module comprises:

The input sub-module is used for inputting each verification video in the verification video set into a pre-trained class identification model to obtain a plurality of class labels corresponding to each verification video, wherein each verification video has an original class label corresponding to the verification video;

The extraction sub-module is used for extracting category labels meeting a first preset screening condition from all category labels corresponding to each verification video to obtain a category label set corresponding to each verification video;

The determining submodule is used for determining all verification videos with the same original category labels in all the verification videos as one verification video group to obtain a plurality of verification video groups;

the screening sub-module is used for screening all category labels meeting a second preset screening condition in all category labels included in the category label sets corresponding to all verification videos in each verification video group to obtain the category label set corresponding to each verification video group;

The establishing sub-module is used for respectively establishing a category label association relation among a plurality of category labels included in the category label set corresponding to each verification video group.

In a second aspect of the present invention, the specific manner of the screening submodule screening all category labels that meet the second preset screening condition in all category labels included in the category label set corresponding to all verification videos in each verification video group to obtain the category label set corresponding to each verification video group is as follows:

As an alternative embodiment, in a second aspect of the present invention, the filtering module includes:

The first filtering sub-module is used for executing a first filtering operation on all videos included in the first video set according to the hash characteristics of the target video and the hash characteristics of each video in the first video set to obtain a filtering result;

And the second filtering sub-module is used for executing a second filtering operation on all videos included in the filtering result according to the hash characteristics of all videos included in the filtering result to obtain a second video set.

In a second aspect of the present invention, according to the hash feature of the target video and the hash feature of each video in the first video set, the first filtering sub-module performs a first filtering operation on all videos included in the first video set, to obtain a filtering result in a specific manner that:

In a second aspect of the present invention, according to an optional implementation manner, the second filtering sub-module performs, according to hash characteristics of all videos included in the filtering result, a second filtering operation on all videos included in the filtering result, so as to obtain a second video set, where a specific manner of obtaining the second video set is:

The third aspect of the present invention discloses another apparatus for controlling a selection of videos to be recommended, the apparatus comprising:

A memory storing executable program code;

a processor coupled to the memory;

the processor invokes the executable program code stored in the memory to execute part or all of the steps in the method for controlling the filtering of the video to be recommended disclosed in the first aspect of the present invention.

The fourth aspect of the present invention discloses a computer storage medium storing computer instructions which, when called, are used for part or all of the steps in the method for controlling the screening of videos to be recommended disclosed in the first aspect of the present invention.

Compared with the prior art, the embodiment of the invention has the following beneficial effects:

In the embodiment of the invention, the target characteristic information of the target video is extracted and the target class label is determined; screening a target multi-category set comprising target category labels from a plurality of multi-category sets, and screening at least one video which meets a preset matching condition with the matching degree of target characteristic information from a video set corresponding to the target multi-category set to obtain a first video set; according to the hash characteristics of the videos, filtering all videos included in the first video set to obtain a second video set, wherein the filtering operation is used for filtering videos with the content overlapping ratio of the video pictures higher than an overlapping ratio threshold value from the first video set; and determining the videos included in the second video set as videos to be recommended of the target videos. Therefore, the method and the device can automatically filter the videos with different picture layouts and higher content overlapping ratio, improve the effectiveness of the screened video to be recommended, and are beneficial to improving the matching degree of the video to be recommended and the user requirement.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic flow chart of a method for controlling screening of video to be recommended according to an embodiment of the present invention.

Fig. 2 is a flow chart of another method for controlling screening of videos to be recommended according to an embodiment of the present invention;

Fig. 3 is a schematic structural diagram of a screening control device based on a video to be recommended according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of another screening control device for video to be recommended according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a screening control device for video to be recommended according to another embodiment of the present invention.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The terms first, second and the like in the description and in the claims and in the above-described figures are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, article, or device that comprises a list of steps or elements is not limited to the list of steps or elements but may, in the alternative, include other steps or elements not expressly listed or inherent to such process, method, article, or device.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

The invention discloses a screening control method and device for videos to be recommended, which can automatically filter videos with different picture layouts and higher content overlapping ratio, improve the effectiveness of the screened videos to be recommended, and are beneficial to improving the matching degree of the videos to be recommended and the requirements of users. The following will describe in detail.

Example one (method side example)

Referring to fig. 1, fig. 1 is a flowchart of a method for controlling a selection of a video to be recommended according to an embodiment of the present invention. The method described in fig. 1 may be applied to a control device, or alternatively, the control device may be a control apparatus, or may be a background server, which is not limited by the embodiment of the present invention. As shown in fig. 1, the method for controlling the filtering of the video to be recommended may include the following operations:

101. The control device extracts the target characteristic information of the target video and determines the target category label of the target video.

In the embodiment of the invention, the target characteristic information of the target video is used for representing the style characteristics of the target video. It should be noted that the target video may be a video currently watched by any user, a video historically watched by any user, or one of the videos previously screened as recommended for any user, which is not limited in the embodiment of the present invention.

In the embodiment of the invention, the target feature information of the target video is extracted by a pre-constructed feature extraction model. If the target video has an original category label, the target category label of the target video can be the original category label, and if the target video does not have the original category label, the target category label of the target video is a prediction category label obtained through the identification of a pre-built category identification model. Alternatively, the process of constructing the feature extraction model and the class identification model may be specifically as follows:

And initializing parameters of a feature extraction model by using a pre-trained three-dimensional machine learning model, wherein the feature extraction model comprises a feature extraction layer and a feature fusion layer. The feature extraction layer mainly comprises a three-dimensional convolution layer and a three-dimensional normalization layer and is used for simultaneously extracting the features of time and space dimensions of the input video; in addition, a full-connection layer is built again and serves as an output layer of category labels of input videos, namely a category identification model (also called a category identification layer), the category identification model outputs m probability values (m represents the number of category labels of the videos), namely for each video, the category identification model outputs probabilities corresponding to the m category labels respectively, and the probability that the video belongs to each category label in the m category labels is represented;

After all layers are initialized by parameters, training is carried out by utilizing a video data set, so that training of a feature extraction model and a category identification model is completed. The class label corresponding to the highest probability output by the class identification model is called as the predicted class label of the video. And the feature vector output by the feature extraction model is the input of the category recognition model, and is obtained by integrating the global information of multiple frames through the feature fusion layer after the feature extraction of the three-dimensional feature extraction layer, so that the feature vector is sufficient to represent the information of the style and the form of the video. Compared with a two-dimensional machine learning model, the three-dimensional machine learning model not only can capture the spatial characteristics of a single frame image, but also can capture the motion characteristics of a video frame sequence in the time dimension, so that the motion information such as picture switching or three-dimensional rotation is extracted, and the correlation degree calculated according to the feature vectors among videos with the same display method can be improved by combining the style characteristics and the motion characteristics, so that the videos can be screened out, and the comprehensiveness and the accuracy of the determined video to be recommended are improved.

It should be noted that, before extracting the target feature information of the target video and the target category label for identifying the target video by using the constructed feature extraction model and the category identification model of the video style, the n video frames extracted from the target video may be subjected to corresponding preprocessing operations, specifically: and uniformly scaling each video frame of the n video frames to a fixed size and carrying out normalization processing. And splicing and fusing the preprocessed video frames on the depth channel, inputting a feature extraction model, so as to obtain feature vectors (namely the target feature information) for representing the video style and the motion information of the target video, and obtaining a target class label (namely a prediction class label) of the target video through a class identification model.

102. The control device screens a target multi-category set comprising target category labels from a plurality of pre-generated multi-category sets, screens at least one video which meets the preset matching condition with the matching degree of the target characteristic information from a video set corresponding to the target multi-category set, and obtains a first video set.

In the embodiment of the invention, each multi-category set may include a plurality of category labels with association relation, a video identifier uniquely corresponding to each video in all videos corresponding to the category labels, and feature information of each video, and further may include a hash feature of each video, where the feature information of each video is used to characterize style features of the video. The control device screens at least one video with the matching degree meeting the preset matching condition from the video sets corresponding to the target multi-category sets to obtain a first video set, and the method can include:

The control device calculates the matching degree of the target characteristic information of the target video and the characteristic information of each video in the video set corresponding to the target multi-category set;

The control device screens all videos with the matching degree larger than or equal to a preset matching degree threshold value from video sets corresponding to the target multi-category sets according to all the calculated matching degrees to obtain a first video set; or alternatively

And the control device screens k videos from video sets corresponding to the target multi-category sets according to the sequence of the matching degree from high to low according to all the calculated matching degrees to obtain a first video set, wherein k is an integer greater than or equal to 1.

Optionally, the matching degree between the videos may be a cosine distance, or may be a euclidean distance, which is not limited by the embodiment of the present invention.

103. And the control device executes filtering operation on all videos included in the first video set according to the determined hash characteristics of the target video and the hash characteristics of each video in the first video set to obtain a second video set.

In the embodiment of the present invention, the filtering operation is used for filtering out, from all videos included in the first video set, videos whose content overlapping ratio of the video frames is higher than a preset overlapping ratio threshold. The hash characteristic of the target video is determined by the following method:

And taking the first j video frames from the extracted n video frames, respectively calculating the hash value of each video frame in the j video frames, and splicing the hash values of the j video frames according to the sequence of the video frames to obtain a spliced hash value as the hash characteristic of the target video, wherein j is an integer greater than or equal to 1 and less than or equal to n, and preferably, the control device can extract the first frame of the target video in 2-4 seconds to obtain 3 video frames (namely j is equal to 3). Because part of the relevant videos are newly added with video title pictures or special effects, larger errors exist when only the video frames of the first two seconds are extracted, the accuracy of the relevant videos obtained by extracting one frame independently for subsequent calculation is lower, the calculated amount is increased when extracting excessive video frames, in addition, because the playing speed of the part of the relevant videos is different and is usually 1.2-1.5 times that of the source videos, if the video frames with relatively later playing time are extracted, the playing pictures of the relevant videos are larger, larger errors exist when the relevant videos obtained by calculation are further caused, and the accuracy is lower.

104. The control device determines videos included in the second video set as videos to be recommended of the target video.

In an alternative embodiment, after the completion of step 104, the control device may further perform the following operations:

The control device acquires a user grade corresponding to the watching user of the target video;

The control device determines a video screening mode matched with the user grade, and screens matched target videos to be recommended from the first video set according to the determined video screening mode;

And the control device recommends the target video to be recommended to a user terminal corresponding to the watching user.

Therefore, the optional embodiment can automatically determine the matched video screening mode according to the user grade after determining the video set to be recommended, and further recommend the corresponding video according to the matched video screening mode, so that the personalized video recommendation based on the user grade is realized.

In another optional embodiment, the control device performs a filtering operation on all the videos included in the first video set to obtain a second video set according to the determined hash feature of the target video and the hash feature of each video in the first video set, and may include:

The control device executes a first filtering operation on all videos included in the first video set according to the hash characteristics of the target video and the hash characteristics of each video in the first video set to obtain a filtering result;

and the control device executes a second filtering operation on all videos included in the filtering result according to the hash characteristics of all videos included in the filtering result to obtain a second video set.

In this optional embodiment, further optionally, the controlling device performs a first filtering operation on all videos included in the first video set according to the hash feature of the target video and the hash feature of each video in the first video set, to obtain a filtering result, and may include:

the control device calculates the Hamming distance between each video in the first video set and the target video according to the Hash characteristics of the target video and the Hash characteristics of each video in the first video set;

the control device filters all videos with the Hamming distance smaller than or equal to a preset Hamming distance threshold value from the first video set to obtain a filtering result.

In this optional embodiment, further optionally, the controlling device performs, according to hash characteristics of all videos included in the filtering result, a second filtering operation on all videos included in the filtering result, to obtain a second video set, and may include:

The control device selects one of the non-selected videos from all the residual videos corresponding to the filtering result, and calculates the Hamming distance between each video except one of the videos and one of the videos in all the residual videos corresponding to the filtering result according to the Hash characteristics of the one of the videos and the Hamming characteristics of each video except one of the videos in all the residual videos corresponding to the filtering result;

The control device judges whether at least one video with the hamming distance smaller than or equal to a preset hamming distance threshold exists in all the remaining videos except one of the videos corresponding to the filtering result;

When the judgment result is no, the control device executes the operation of selecting one of the non-selected videos from all the residual videos corresponding to the filtration result, calculating the hamming distance between each of the all the residual videos corresponding to the filtration result and one of the videos according to the hash characteristics of the one of the videos and the hash characteristics of each of the other videos in all the residual videos corresponding to the filtration result, and executing the operation of judging whether at least one of the videos with the hamming distance smaller than or equal to the preset hamming distance threshold exists in all the residual videos with the exception of the one of the videos;

When the judgment result is yes, the control device filters all videos of which the Hamming distance is smaller than or equal to a preset Hamming distance threshold value from all the videos except one of the videos corresponding to the filtering result, so as to update all the remaining videos corresponding to the filtering result, and executes the operation of selecting one of the videos which is not selected from all the remaining videos corresponding to the filtering result, according to the hash characteristics of one of the videos and the hash characteristics of each video except one of the videos in all the remaining videos corresponding to the filtering result, calculating the Hamming distance between each video except one of the videos and one of the videos in all the remaining videos corresponding to the filtering result, and executing the operation of judging whether the Hamming distance between each video except one of the videos and at least one of the videos in all the remaining videos corresponding to the filtering result is smaller than or equal to the preset Hamming distance threshold value;

When only one unselected video exists in all the residual videos corresponding to the filtering result, the control device determines all the residual videos corresponding to the filtering result as a second video set.

Therefore, in the optional embodiment, the high-dimensional features of the video frames corresponding to the video can be mapped to the low-dimensional space through the hash features of the video, and as the video frames with the same video playing frames (i.e. playing contents) or high coincidence degree have the hash features with high similarity, the video with high playing frame repetition degree in the preliminarily determined video set is filtered through the hash features of the video, so that the effectiveness of the determined video set to be recommended is improved.

Therefore, the embodiment of the invention can automatically filter the videos with different picture layouts and higher content overlapping ratio, improves the effectiveness of the screened video to be recommended, and is beneficial to improving the matching degree of the video to be recommended and the user requirement. In addition, the matched video screening mode can be automatically determined according to the user grade after the video set to be recommended is determined, and then the corresponding video is recommended according to the matched video screening mode, so that the personalized video recommendation based on the user grade is realized. In addition, the high-dimensional features of the video frames corresponding to the videos can be mapped to the low-dimensional space through the hash features of the videos, and as the video frames with the same video playing pictures (namely playing contents) or high coincidence degree have the hash features with high similarity, the video with high playing picture repetition degree in the preliminarily determined video set is filtered through the hash features of the videos, so that the effectiveness of the determined video set to be recommended is improved.

Example two (method side example)

Referring to fig. 2, fig. 2 is a flowchart illustrating another method for controlling a filtering of video to be recommended according to an embodiment of the present invention. The method described in fig. 2 may be applied to a control device, or alternatively, the control device may be a control apparatus, or may be a background server, which is not limited by the embodiment of the present invention. As shown in fig. 2, the method for controlling the filtering of the video to be recommended may include the following operations:

201. the control device generates a category label association relationship.

202. The control device stores the video identification, the characteristic information, the hash characteristic and the category label of each video in the original video set to be recommended as an information set corresponding to the video.

In the embodiment of the invention, if a certain video of the original video set has an original category label, the category label included in the information set corresponding to the video is the original category label of the video, and if the video does not have the original category label, the category label included in the information set corresponding to the video is the prediction category label of the video.

203. And the control device executes concentrated storage operation on the information sets of all the category labels with the association relationship according to the predetermined category label association relationship to obtain a plurality of multi-category sets.

204. The control device extracts the target characteristic information of the target video and determines the target category label of the target video.

In the embodiment of the invention, the target characteristic information of the target video is used for representing the style characteristics of the target video.

205. The control device screens a target multi-category set comprising target category labels from the generated multiple multi-category sets, screens at least one video which meets the preset matching condition with the matching degree of the target characteristic information from a video set corresponding to the target multi-category set, and obtains a first video set.

206. And the control device executes filtering operation on all videos included in the first video set according to the determined hash characteristics of the target video and the hash characteristics of each video in the first video set to obtain a second video set.

In the embodiment of the present invention, the filtering operation is used for filtering out, from all videos included in the first video set, videos whose content overlapping ratio of the video frames is higher than a preset overlapping ratio threshold.

207. The control device determines videos included in the second video set as videos to be recommended of the target video.

In the embodiment of the present invention, the detailed descriptions of step 204 to step 207 are referred to in the first embodiment for the detailed descriptions of step 101 to step 104, and the detailed descriptions of the embodiment of the present invention are omitted.

Therefore, the method described by the embodiment of the invention can automatically filter the video with different picture layouts and higher content overlapping ratio, improves the effectiveness of the screened video to be recommended, and is beneficial to improving the matching degree of the video to be recommended and the user requirement. In addition, the related videos of different category labels can be screened out in a mode of generating the category label association relation, and accuracy of the screened related videos is improved.

In an alternative embodiment, the control device generates a category label association relationship, which may include:

The control device inputs each verification video in the verification video set into a pre-trained class identification model to obtain a plurality of class labels corresponding to each verification video, wherein each verification video has an original class label corresponding to the verification video;

the control device extracts category labels meeting a first preset screening condition from all category labels corresponding to each verification video to obtain a category label set corresponding to each verification video;

the control device determines all verification videos with the same original category labels in all verification videos as one verification video group to obtain a plurality of verification video groups;

the control device screens all category labels meeting a second preset screening condition in all category labels included in the category label set corresponding to all verification videos in each verification video group to obtain a category label set corresponding to each verification video group;

The control device respectively establishes a category label association relationship among a plurality of category labels included in the category label set corresponding to each verification video group.

Further optionally, the controlling device screens all category labels meeting the second preset screening condition from all category labels included in the category label set corresponding to all verification videos in each verification video group, to obtain the category label set corresponding to each verification video group, and may include:

In this optional embodiment, further optionally, the controlling means counts the number of labels of each class label in the class label set corresponding to all the verification videos in the verification video group, and may include:

the control device counts the accumulated times of the occurrence of the same category labels in the category label set corresponding to each verification video in the verification video group to obtain the label number corresponding to each category label.

For example, assuming that a certain verification video group includes 3 verification videos with a original category label a, the category label set corresponding to the first verification video is a, b, and c, the category label set corresponding to the second verification video is a, b, and d, and the category label set corresponding to the third verification video is a, b, and e, the number of labels corresponding to each category label counted is: the number of marks corresponding to the category label a is 3, the number of marks corresponding to the category label b is 3, the number of marks corresponding to the category label c, the category label d and the category label e is 1, and if the preset number of marks threshold is 2, the category label set corresponding to the verification video group comprises a category label a and a category label b with the number of marks of 3.

In the alternative embodiment, because videos of different category labels may have a certain correlation, the correlation between the category labels of the related videos is built through the category identification model trained in advance, so that the comprehensiveness and the accuracy of the determined target multi-category set can be improved, the comprehensiveness and the accuracy of videos included in the video set to be recommended which is determined later are improved, and the situation that the related videos cannot be recommended to users due to different category labels is reduced.

Therefore, the embodiment of the invention can automatically filter the videos with different picture layouts and higher content overlapping ratio, improves the effectiveness of the screened videos to be recommended, is beneficial to improving the matching degree of the videos to be recommended and the requirements of users, and can uniformly store the information set of each video according to the category labels, thereby being beneficial to reducing the irrelevant characteristic range during characteristic retrieval and further accelerating the retrieval speed. In addition, the information set of the category labels with the association relationship can be stored in a centralized manner through the association relationship among the category labels, so that the effective range of feature retrieval is enlarged, and when the original category labels do not exist in the video, the category labels predicted by the pre-trained category recognition model can be retrieved, so that the retrieval effectiveness is improved.

Example III (device side example)

Referring to fig. 3, fig. 3 is a schematic structural diagram of a device for screening and controlling video to be recommended according to an embodiment of the present invention. The apparatus described in fig. 3 may be specifically applied to a control device or a background server, and embodiments of the present invention are not limited thereto. As shown in fig. 3, the apparatus for screening and controlling video to be recommended may include:

the extracting module 301 is configured to extract target feature information of a target video, where the target feature information is used to characterize style features of the target video.

A determining module 302, configured to determine a target category label of the target video.

The first screening module 303 is configured to screen a target multi-category set including a target category label from a plurality of pre-generated multi-category sets.

The second screening module 304 is configured to screen at least one video that has a matching degree with the target feature information that meets a preset matching condition from the video sets corresponding to the target multi-category set, so as to obtain a first video set.

The filtering module 305 is configured to perform a filtering operation on all videos included in the first video set according to the determined hash feature of the target video and the hash feature of each video in the first video set to obtain a second video set, where the filtering operation is used to filter, from all videos included in the first video set, videos whose content overlapping ratio of the video frames is higher than a preset overlapping ratio threshold.

The determining module 302 is further configured to determine a video included in the second video set as a video to be recommended of the target video.

Therefore, the device described in fig. 3 can automatically filter the video with different picture layouts and higher content overlapping ratio, so as to improve the effectiveness of the screened video to be recommended, and be beneficial to improving the matching degree of the video to be recommended and the user requirement.

In an alternative embodiment, as shown in fig. 4, the apparatus may further include:

the multi-category set generating module 306 is configured to store, as an information set corresponding to each video in the original video set to be recommended, a video identifier, feature information, a hash feature, and a category tag of the video before the first screening module 303 screens a target multi-category set including a target category tag from a plurality of pre-generated multi-category sets; and according to the pre-determined category label association relation, performing centralized storage operation on the information sets of all category labels with the association relation to obtain a plurality of multi-category sets.

Each multi-category set comprises a plurality of category labels with association relations and contents in information sets corresponding to the corresponding videos.

Therefore, the device described in fig. 4 can also screen out related videos of different category labels by generating the category label association relationship, which is beneficial to improving the accuracy of the screened out related videos.

In another alternative embodiment, as shown in fig. 4, the apparatus may further include:

The association relation generating module 307 is configured to generate a category label association relation.

Further optionally, the association generating module 307 includes:

And an input submodule 3071, configured to input each verification video in the verification video set into a pre-trained class identification model, and obtain a plurality of class labels corresponding to each verification video, where each verification video has an original class label corresponding to each verification video.

The extracting submodule 3072 is configured to extract category labels that satisfy the first preset screening condition from all category labels corresponding to each verification video, so as to obtain a category label set corresponding to each verification video.

The determining submodule 3073 is configured to determine all verification videos with the same original category labels in all verification videos as one verification video group, so as to obtain a plurality of verification video groups.

And a screening submodule 3074, configured to screen all category labels meeting the second preset screening condition from all category labels included in the category label sets corresponding to all verification videos in each verification video group, so as to obtain the category label set corresponding to each verification video group.

The establishing submodule 3075 is used for respectively establishing a category label association relationship among a plurality of category labels included in the category label set corresponding to each verification video group.

Further optionally, the specific manner of the screening submodule 3074 for screening all category labels meeting the second preset screening condition in all category labels included in the category label set corresponding to all the verification videos in each verification video group to obtain the category label set corresponding to each verification video group may be:

Therefore, the device described in fig. 4 can also store the information set of the category label with the association relationship in a centralized manner through the association relationship between the category labels, which is favorable for expanding the effective range of feature retrieval, and when the video does not have the original category label, the device can retrieve the category label predicted by the pre-trained category recognition model, thereby improving the retrieval effectiveness.

In yet another alternative embodiment, as shown in fig. 4, the filtering module 305 may include:

The first filtering submodule 3051 is used for executing a first filtering operation on all videos included in the first video set according to the hash characteristics of the target video and the hash characteristics of each video in the first video set to obtain a filtering result;

And the second filtering sub-module 3052 is configured to perform a second filtering operation on all videos included in the filtering result according to the hash characteristics of all videos included in the filtering result, so as to obtain a second video set.

Further optionally, the specific manner of performing, by the first filtering submodule 3051, the first filtering operation on all the videos included in the first video set according to the hash feature of the target video and the hash feature of each video in the first video set to obtain the filtering result may be:

Further optionally, the second filtering sub-module 3052 performs a second filtering operation on all videos included in the filtering result according to hash features of all videos included in the filtering result, and obtaining the second video set may include:

selecting one of the non-selected videos from all the residual videos corresponding to the filtering result, and calculating the Hamming distance between each video except one of the videos in all the residual videos corresponding to the filtering result and one of the videos according to the Hash characteristics of the one of the videos and the Hash characteristics of each video except one of the videos in all the residual videos corresponding to the filtering result;

judging whether at least one video with the hamming distance smaller than or equal to a preset hamming distance threshold exists in all the remaining videos except one of the videos corresponding to the filtering result;

When the judgment result is no, triggering and executing the operation of selecting one of the non-selected videos from all the residual videos corresponding to the filtering result, calculating the hamming distance between each of the all the residual videos corresponding to the filtering result except one of the videos and one of the videos according to the hash characteristics of the one of the videos and the hash characteristics of each of the all the residual videos except one of the videos corresponding to the filtering result, and triggering and executing the operation of judging whether at least one of the videos with the hamming distance smaller than or equal to the preset hamming distance threshold exists in all the residual videos except one of the videos corresponding to the filtering result;

When the judgment result is yes, filtering all videos of which the Hamming distance is smaller than or equal to a preset Hamming distance threshold value from all the remaining videos corresponding to the filtering result except one of the videos, updating all the remaining videos corresponding to the filtering result, triggering and executing the operation of selecting one of the videos which is not selected from all the remaining videos corresponding to the filtering result, calculating the Hamming distance between each video except one of the videos and one of the videos in all the remaining videos corresponding to the filtering result according to the Hamming characteristics of the one of the videos and the Hamming characteristics of each video except one of the videos in all the remaining videos corresponding to the filtering result, and triggering and executing the operation of judging whether the Hamming distance between each video except one of the videos and at least one of the videos is smaller than or equal to the preset Hamming distance threshold value;

As can be seen, the device described in fig. 4 can map the high-dimensional features of the video frames corresponding to the video to the low-dimensional space through the hash features of the video, and since the video frames with the same video playing frames (i.e. playing contents) or high overlapping degree have the hash features with high similarity, the filtering of the video with high playing frame repeatability in the preliminarily determined video set is realized through the hash features of the video, which is beneficial to improving the effectiveness of the determined video set to be recommended.

Example IV

Referring to fig. 5, fig. 5 is a schematic structural diagram of a screening control device for video to be recommended according to an embodiment of the present invention. The apparatus described in fig. 5 may be specifically applied to a control device or a background server, and embodiments of the present invention are not limited thereto. As shown in fig. 5, the apparatus may include:

A memory 401 storing executable program codes;

a processor 402 coupled with the memory 401;

The processor 402 invokes executable program codes stored in the memory 401, for executing some or all of the steps in the method for controlling filtering of video to be recommended described in the first or second embodiment.

Example five

The embodiment of the invention discloses a computer storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute part or all of the steps in the method for controlling the screening of videos to be recommended described in the first embodiment or the second embodiment.

Example six

An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps in the method for controlling filtering of videos to be recommended described in the first embodiment or the second embodiment.

The apparatus embodiments described above are merely illustrative, wherein the modules illustrated as separate components may or may not be physically separate, and the components shown as modules may or may not be physical, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

From the above detailed description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course by means of hardware. Based on such understanding, the foregoing technical solutions may be embodied essentially or in part in the form of a software product that may be stored in a computer-readable storage medium including Read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), programmable Read-Only Memory (Programmable Read-Only Memory, PROM), erasable programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM), one-time programmable Read-Only Memory (OTPROM), electrically erasable programmable Read-Only Memory (EEPROM), compact disc Read-Only Memory (Compact Disc Read-Only Memory, CD-ROM) or other optical disc Memory, magnetic disc Memory, tape Memory, or any other medium that can be used for computer-readable carrying or storing data.

Finally, it should be noted that: the embodiment of the invention discloses a method and a device for screening and controlling video to be recommended, which are disclosed by the embodiment of the invention only as a preferred embodiment of the invention, and are only used for illustrating the technical scheme of the invention, but not limiting the technical scheme; although the invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art will understand that; the technical scheme recorded in the various embodiments can be modified or part of technical features in the technical scheme can be replaced equivalently; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.

Claims

1. The method for screening and controlling the video to be recommended is characterized by comprising the following steps:

Screening a target multi-category set comprising the target category label from a plurality of pre-generated multi-category sets, and screening at least one video which meets a preset matching condition with the matching degree of the target characteristic information from a video set corresponding to the target multi-category set to obtain a first video set; the multi-category sets are obtained by performing centralized storage operation on information sets to which all category labels with association relationships belong according to the category label association relationships determined in advance;

determining videos included in the second video set as videos to be recommended of the target video;

Wherein, the category label association relation is determined by the following modes: inputting each verification video in a verification video set into a pre-trained class identification model to obtain a plurality of class labels corresponding to each verification video, wherein each verification video has an original class label corresponding to the verification video; extracting category labels meeting a first preset screening condition from all category labels corresponding to each verification video to obtain a category label set corresponding to each verification video; determining all verification videos with the same original category labels in all the verification videos as one verification video group to obtain a plurality of verification video groups; screening all category labels meeting a second preset screening condition in all category labels included in the category label set corresponding to all verification videos in each verification video group to obtain a category label set corresponding to each verification video group; and respectively establishing a category label association relationship among a plurality of category labels included in the category label set corresponding to each verification video group.

2. The method according to claim 1, wherein before the selecting a target multi-category set including the target category label from a plurality of pre-generated multi-category sets, the method further comprises:

3. The method for screening and controlling videos to be recommended according to claim 2, wherein the method further comprises, before performing a centralized storage operation on information sets to which all category labels having an association relationship belong according to a predetermined category label association relationship to obtain a plurality of multi-category sets:

And generating a category label association relation.

4. The method for screening and controlling videos to be recommended according to claim 3, wherein the step of screening all category labels meeting a second preset screening condition from all category labels included in the category label set corresponding to all verification videos in each verification video group to obtain the category label set corresponding to each verification video group includes:

5. The method for controlling filtering of videos to be recommended according to any one of claims 1 to 4, wherein the performing filtering operation on all videos included in the first video set to obtain a second video set according to the determined hash feature of the target video and the hash feature of each video in the first video set includes:

6. The method for filtering and controlling videos to be recommended according to claim 5, wherein the performing a first filtering operation on all the videos included in the first video set according to the hash feature of the target video and the hash feature of each video in the first video set to obtain a filtering result includes:

7. The method for screening and controlling videos to be recommended according to claim 6, wherein the performing a second filtering operation on all videos included in the filtering result according to hash features of all videos included in the filtering result to obtain a second video set includes:

8. A screening control device for video to be recommended, the device comprising:

The first screening module is used for screening a target multi-category set comprising the target category label from a plurality of pre-generated multi-category sets; the multi-category sets are obtained by performing centralized storage operation on information sets to which all category labels with association relationships belong according to the category label association relationships determined in advance;

the determining module is further configured to determine a video included in the second video set as a video to be recommended of the target video;

9. A screening control device for video to be recommended, the device comprising:

A memory storing executable program code;

a processor coupled to the memory;

The processor invokes the executable program code stored in the memory to perform the method of screening control of video to be recommended as claimed in any one of claims 1 to 7.

10. A computer storage medium storing computer instructions which, when invoked, are adapted to perform the method of screening control of videos to be recommended according to any one of claims 1-7.