CN112749296A

CN112749296A - Video recommendation method and device, server and storage medium

Info

Publication number: CN112749296A
Application number: CN201911056969.XA
Authority: CN
Inventors: 宋林枫
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2019-10-31
Filing date: 2019-10-31
Publication date: 2021-05-04
Anticipated expiration: 2039-10-31
Also published as: CN112749296B

Abstract

The disclosure relates to a video recommendation method, a video recommendation device, a server and a storage medium, relating to the technical field of computers and used for solving the problem of low recall rate of video retrieval in the related technology, wherein the method comprises the following steps: acquiring a corresponding account feature vector according to a received video recommendation request, wherein the video recommendation request carries information of an account of a video to be recommended, and the account feature vector corresponds to the information of the account; acquiring a constructed video recommendation index, wherein the video recommendation index comprises M layers, each layer comprises a preset feature vector, the Nth layer of the M layers comprises a preset feature vector contained in the (N-1) th layer, M is a natural number, and N is a natural number which is greater than 1 and less than M; comparing the similarity of the account characteristic vector serving as an input with a preset characteristic vector included in the video recommendation index layer by layer to obtain a comparison result; recommending videos to the account according to the comparison result output from the Mth layer.

Description

Video recommendation method and device, server and storage medium

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a video recommendation method, an apparatus, a server, and a storage medium.

Background

With the popularization of the internet, more and more users watch videos, and in order to meet the personalized requirements of the users, a video player usually adopts a real-time video recommendation system to recommend videos which may be interested by the users to the users.

At present, a tree-based index structure is more often adopted in a real-time video recommendation system, for example, a binary tree index structure established based on an Approximate neighbor algorithm (Annoy), where the binary tree index structure is a process of continuously traversing from a root node to a leaf node, and whether a binary tree traversal process goes to a left child node or a right child node of the middle node is determined by performing correlation calculation on each middle node (segmentation hyperplane related information) of a binary tree and an inquiry data node. In the tree-based index structure, the number of data nodes of leaf nodes is limited, and the condition that the number of the data nodes finally falling to the leaf nodes is smaller than the number of the data nodes needing to be recommended or two similar data nodes are divided into different branches of a binary tree may occur in the query process, so that the video retrieval recall rate is low, the quality of video recommendation service is influenced, and the user experience is not good.

Disclosure of Invention

The present disclosure provides a video recommendation method, apparatus, server and storage medium to at least solve the problem of low recall rate of video retrieval in the related art.

The technical scheme of the disclosure is as follows:

in a first aspect of the embodiments of the present disclosure, a video recommendation method is provided, including:

acquiring a corresponding account feature vector according to a received video recommendation request, wherein the video recommendation request carries information of an account of a video to be recommended, and the account feature vector corresponds to the information of the account;

acquiring a constructed video recommendation index, wherein the video recommendation index comprises M layers, each layer comprises a preset feature vector, the Nth layer of the M layers comprises a preset feature vector contained in the (N-1) th layer, M is a natural number, and N is a natural number which is greater than 1 and less than M;

comparing the similarity of the account feature vector serving as an input with a preset feature vector included in the video recommendation index layer by layer to obtain a comparison result, wherein the input aiming at the first layer of the video recommendation index is the account feature vector, and the input aiming at the Nth layer is the comparison result of the (N-1) th layer;

recommending videos to the account according to the comparison result output from the Mth layer.

In one possible design, the video recommendation index is constructed as follows:

acquiring a preset characteristic vector set;

constructing an index level mapping table based on the preset feature vector set, wherein the index level mapping table records the number of index layers, the number of feature vectors distributed in each layer and information of the layer to which the feature vectors belong;

and distributing preset feature vectors to each layer according to the index level mapping table to obtain the video recommendation index.

In one possible design, after obtaining the set of preset feature vectors, the method further includes:

updating the preset characteristic vector set at preset time intervals;

and updating the index level mapping table based on the updated preset feature vector set.

In one possible design, updating the set of preset feature vectors includes:

deleting the preset feature vectors which are not changed after a preset time length in the preset feature vector set; and/or the presence of a gas in the gas,

and adding a new preset feature vector to the preset feature vector set.

In one possible design, constructing an index hierarchy mapping table based on the preset feature vector set includes:

determining the number of layers included in the index level mapping table and the number of preset feature vectors distributed in each layer according to a preset increment rate;

selecting a corresponding number of preset feature vectors from the preset feature vector set according to the number of the preset feature vectors distributed in each layer;

and determining the distribution position of the preset feature vectors in each layer in the index hierarchical mapping table according to the similarity height relationship between the preset feature vectors included in each layer.

In one possible design, the selecting a corresponding number of predetermined eigenvectors includes:

and randomly selecting from the preset feature vector set, or sequentially selecting according to the arrangement sequence of each preset feature vector in the preset feature vector set.

In a possible design, if a corresponding number of preset feature vectors are sequentially selected according to an arrangement order of each preset feature vector in the preset feature vector set, before reconstructing the video recommendation index after updating the preset feature vector set, the method further includes:

and randomly scattering the current arrangement sequence of each preset feature vector in the updated preset feature vector set.

In one possible design, recommending videos to the account according to the comparison result output from the mth layer includes:

determining a preset feature vector of which the similarity with the account feature vector meets a preset condition from a comparison result output by the Mth layer;

and recommending the video corresponding to the determined preset characteristic vector to the account as a target video.

In one possible design, the preset condition includes one of the following conditions: the similarity is greater than a first predetermined threshold; the similarity is ordered at the top K bits, and K is a natural number greater than 1.

In one possible design, when the preset condition is that the similarity ranks K top, determining the comparison result of each layer of the video recommendation index includes:

comparing the distance between the characteristic vector input by each layer and the preset characteristic vector distributed in the layer;

and determining the preset feature vector closest to the input feature vector as the preset feature vector with the highest similarity to the input feature vector, and taking the preset feature vector with the highest similarity as the comparison result of each layer.

In one possible design, after determining the target video recommended to the account according to the comparison result output by the mth layer, before performing video recommendation, the method further includes: processing the target video recommended to the account according to one or the combination of the following conditions:

deleting the video with repeated content in the target video; or,

deleting videos of which the video definition is lower than a second preset threshold value in the target videos; or,

and compressing the target video.

According to a second aspect of the embodiments of the present disclosure, there is provided a video recommendation apparatus including:

the video recommendation method comprises a receiving unit, a recommending unit and a recommending unit, wherein the receiving unit is configured to execute acquiring a corresponding account feature vector according to a received video recommendation request, the video recommendation request carries information of an account of a video to be recommended, and the account feature vector corresponds to the information of the account;

the video recommendation method comprises an acquisition unit, a calculation unit and a display unit, wherein the acquisition unit is configured to execute acquisition of a constructed video recommendation index, the video recommendation index comprises M layers, each layer comprises a preset feature vector, the Nth layer of the M layers comprises a preset feature vector contained in the (N-1) th layer, M is a natural number, and N is a natural number which is greater than 1 and less than M;

the determining unit is configured to perform similarity comparison between the account feature vector as an input and a preset feature vector included in the video recommendation index layer by layer to obtain a comparison result, wherein the input of a first layer of the video recommendation index is the account feature vector, and the input of an Nth layer is the comparison result of an (N-1) th layer;

a recommending unit configured to perform recommending a video to the account according to the comparison result output from the Mth layer.

In one possible design, the apparatus further includes a construction unit, and the video recommendation index is constructed by the construction unit in the following manner, and the construction unit is configured to perform:

acquiring a preset characteristic vector set;

In one possible design, the apparatus further includes an updating unit configured to perform:

updating the preset characteristic vector set at preset time intervals;

In one possible design, the update unit is specifically configured to perform:

and adding a new preset feature vector to the preset feature vector set.

In one possible design, the building unit is further configured to perform:

and randomly selecting a corresponding number of preset eigenvectors from the preset eigenvector set, or sequentially selecting a corresponding number of preset eigenvectors according to the arrangement sequence of each preset eigenvector in the preset eigenvector set.

In one possible design, the apparatus further includes a sorting unit configured to perform:

and randomly scattering the current arrangement sequence of each preset feature vector in the updated preset feature vector set when the corresponding number of preset feature vectors is sequentially selected according to the arrangement sequence of each preset feature vector in the preset feature vector set.

In one possible design, the recommending unit is further configured to perform:

In one possible design, the determining unit is specifically configured to perform:

when the preset condition is that the similarity is ranked at the top K, comparing the distance between the feature vector input by each layer and the preset feature vectors distributed in the layer;

In one possible design, the apparatus further includes a processing unit configured to process the target video recommended to the account according to one or a combination of the following conditions:

deleting the video with repeated content in the target video; or,

and compressing the target video.

According to a third aspect of the embodiments of the present disclosure, there is provided a video recommendation server, including:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the first aspect of the embodiments of the present disclosure described above and any method to which the first aspect relates.

According to a fourth aspect of the embodiments of the present disclosure, there is provided a storage medium, where instructions, when executed by a processor of a video recommendation server, enable the video recommendation server to perform the first aspect of the embodiments of the present disclosure and any of the methods that the first aspect relates to may be performed.

According to a fifth aspect of the embodiments of the present disclosure, there is provided a computer program product, which, when run on a video recommendation server, causes the video recommendation server to perform a method for implementing the first aspect of the embodiments of the present disclosure and any possible method related to the first aspect.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

in the embodiment of the disclosure, a layered video recommendation index is adopted for video recommendation, after a video recommendation request is received, a corresponding account feature vector can be obtained according to the video recommendation request, wherein the video recommendation request carries information of an account of a video to be recommended, the account feature vector corresponds to the account information of the video to be recommended, the obtained account feature vector can be input into the video recommendation index, similarity comparison is performed on the account feature vector and a preset feature vector included in the video recommendation index layer by layer according to an index mode of multiplexing comparison results of adjacent low layers at a high layer, a comparison result with similarity meeting a preset condition is obtained, and the video is recommended to the account of the video to be recommended according to the comparison result output by the highest layer. Therefore, compared with the existing tree-based video recommendation index, the embodiment of the disclosure provides a new video recommendation index adopting a layer structure, because the preset feature vectors in adjacent low layers are included in the high layers in the layer structure, and the comparison result of multiplexing the adjacent low layers by the high layers is adopted during retrieval, the calculation amount can be effectively reduced, and the lock conflict times can be reduced, so that the recall rate of video retrieval can be improved, the quality of video recommendation service can be improved, and the user experience can be improved.

Furthermore, in the embodiment of the disclosure, preset feature vectors in the video recommendation index may be scanned at regular time, feature vectors that have not been updated for a long time are deleted, a new feature vector is added to obtain a new preset feature vector set, and after the identification sequence of each preset feature vector in the new preset feature vector set is broken, a new video recommendation index is reconstructed, so that the problem that the native HNSW does not support feature vector updating and deletion can be solved, and then the recall rate of video retrieval can be improved, so that the recommended video better meets the user requirements, and therefore, the user experience can be further improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.

FIG. 1 is a diagram illustrating an application scenario in accordance with an exemplary embodiment;

FIG. 2 is a flow diagram illustrating a video recommendation method according to an exemplary embodiment;

FIG. 3 is a schematic diagram illustrating a video recommendation index hierarchy relationship in accordance with an illustrative embodiment;

FIG. 4a is a schematic diagram of a video recommendation device according to an exemplary embodiment;

FIG. 4b is another schematic diagram of a video recommendation device, shown in accordance with an exemplary embodiment;

FIG. 5 is a schematic diagram illustrating the structure of a video recommendation server in accordance with an exemplary embodiment;

fig. 6 is another structural diagram of a video recommendation server according to an example embodiment.

Detailed Description

In order to make the technical solutions disclosed in the embodiments of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first", "second", and the like in the description and the claims of the embodiments disclosed in the present disclosure and the drawings described above are used for distinguishing similar objects and not necessarily for describing a particular order or sequence. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

In addition, the term "and/or" herein is only one kind of association relationship describing an associated object, and means that there may be three kinds of relationships, for example, a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in this document generally indicates that the preceding and following related objects are in an "or" relationship unless otherwise specified.

In order to facilitate understanding of the technical solutions provided by the embodiments of the present disclosure, some key terms used by the embodiments of the present disclosure are explained first:

HNSW: (hierarchical Navigable Small World) HNSW is a graph-index based approach proposed by Yury.A.Markov, which builds up a multi-level structure step by step with elements stored in an upper level nested in a lower level, and upon retrieval using the algorithm, the algorithm first traverses the elements of the uppermost level until a locally smaller value is determined, after which the search switches to the lower level, restarts traversing the elements of the lower level according to the locally smaller value in the previous level, and the process repeats.

Word vector characteristics: or called Embedding feature, the object video of the Embedding feature in the embodiment of the present disclosure is used to describe a video. The description idea of the word vector features is to convert words represented by natural language into vectors or matrix forms that can be understood by a computer, and the extraction of the word vector features may be performed by a deep learning model, for example, a Convolutional Neural Network (CNN) model, a Long Short-Term Memory Network (LSTM) model, a Recurrent Neural Network (RNN) model, or a Gated CNN (G-CNN) model, and of course, other possible deep learning models may be used for extraction, which is not limited in this disclosure.

Video retrieval recall ratio: the ratio of the retrieved video to all videos stored in the video library measures the recall ratio of the retrieval system, e.g., 500 videos in a database, of which 50 videos meet the definition. The system retrieves 75 videos, but only 45 are actually in line with the definition. Then: the recall rate is 45/50-90%.

For ease of understanding, the technical background of the embodiments of the present disclosure is described below.

As described above, the current video recommendation system has a problem of low recall rate based on the Annoy tree index structure, so that the recommended video quality is poor, and the user experience is affected.

In view of this, the inventor of the present disclosure proposes a scheme for video recommendation, in which a graph-based indexing method is adopted to construct a layered video recommendation index, and then a video is retrieved by using the layered video recommendation index, and video recommendation is performed according to a retrieval result.

In this scheme, because layered video recommendation indexes are adopted in the embodiment of the present disclosure, account feature vectors in each layer may be sequentially linked together according to the similarity level, so that a situation that two feature vectors with higher similarity levels are divided into different branches based on a tree-shaped index structure does not occur, and further, each layer of the video recommendation indexes may be sequentially traversed from the lowest layer to the highest layer according to an index manner in which a comparison result of adjacent low layers is multiplexed by the higher layer, similarity comparison is performed between a feature vector input in each layer and preset feature vectors distributed in the layer, the preset feature vectors with similarity satisfying a preset condition are used as the comparison result, and a video is recommended to an account according to the obtained comparison result output by the highest layer. Therefore, compared with the existing tree-based video recommendation index, the embodiment of the present disclosure provides a new video recommendation index adopting a layer structure, since the preset feature vectors in the adjacent lower layers are included in the higher layer in the layer structure, and the comparison result of multiplexing the adjacent lower layers by the higher layer is adopted during retrieval, the amount of computation can be effectively reduced, and the number of times of lock conflicts can be reduced, so that the recall rate of video retrieval can be improved, the quality of video recommendation service can be improved, and the user experience can be improved.

Some simple descriptions are given below to application scenarios to which the technical solution of the embodiment of the present disclosure can be applied, and it should be noted that the application scenarios described below are only used for describing the embodiment of the present disclosure and are not limited. In specific implementation, the technical scheme provided by the embodiment of the disclosure can be flexibly applied according to actual needs.

Referring to an application scenario schematic diagram shown in fig. 1, the scenario includes a terminal device 101 and a server 102, where the terminal device 101 may be a terminal device such as a smart phone, a tablet computer, a computer, various wearable devices, and a vehicle-mounted device, the server 102 may be a server, or may be a server cluster or a cloud computing center formed by a plurality of servers, the terminal device 101 communicates with the server 102 through a network, and the network may be any one of communication networks such as a local area network, a wide area network, and a mobile internet.

The terminal device 101 has installed therein an application program that can play a video, where the application program may be, for example, a video application or a news application, and when the application program is the video application, the application program may directly display a recommended video on a display page in the video application, and when the video is selected, the video may be played through the display page of the video application, or the application program may be, for example, a browser, and a page of a video website may be opened in the browser to display the recommended video on the display page.

The server 102 may store the learned account feature vector, the trained model parameter, the label information of the user and the video, and the like, where the account feature vector includes a video feature vector of a video accessed by the account or an account attribute feature vector representing information of interest, age, occupation, gender, and the like of the user. The server 102 is a backend server that can be a video playing application installed in the terminal device 101, which is specifically illustrated in fig. 1 by way of example, but of course, the server 102 may also be a dedicated server for video recommendation, for example, the server 102 is configured to calculate videos recommended for each account, then send recommendation data and the like to a backend server of a video website, and then perform video push by the video website.

It should be noted that the above-mentioned application scenarios are merely illustrated for the convenience of understanding the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. Rather, the disclosed embodiments may be applied to any scenario where applicable.

To further illustrate the technical solutions provided by the embodiments of the present disclosure, the following detailed description is made with reference to the accompanying drawings and the specific embodiments. Although the disclosed embodiments provide method steps as shown in the following embodiments or figures, more or fewer steps may be included in the methods based on conventional or non-inventive efforts. In steps where no necessary causal relationship exists logically, the order of execution of the steps is not limited to that provided by the disclosed embodiments. The method can be executed in sequence or in parallel according to the method shown in the embodiment or the figures when the method is executed in an actual processing procedure or a device (for example, a parallel processor or an application environment of multi-thread processing).

Fig. 2 is a flowchart illustrating a video recommendation method according to an exemplary embodiment, which may be applied to the terminal device 101 shown in fig. 1, as shown in fig. 2, where the method shown in fig. 2 includes the following steps.

In step 201, according to a received video recommendation request, a corresponding account feature vector is obtained, where the video recommendation request carries information of an account of a video to be recommended, and the account feature vector corresponds to the information of the account.

The video recommendation request in the embodiment of the present disclosure may be generated by a user inputting a keyword for searching a video in a video application of the terminal device 101 shown in fig. 1 or a search bar of a video website in a browser, and then the server 102 shown in fig. 1 may receive the video recommendation request sent by an account corresponding to the video application of the terminal device 101; of course, the video recommendation request may also be a video recommendation request generated by an action such as a pull-down or a slide when the user watches a video on the terminal device 101 shown in fig. 1 through the video application account, and then the server 102 may receive the video recommendation request. For convenience of description, in the embodiments of the present disclosure, a video recommendation in a video application is specifically described as an example below.

In the embodiment of the present disclosure, the information of the account includes a video accessed by the account and user attribute information of interest, age, occupation, gender, and the like of a user corresponding to the account, and the corresponding account feature vector may be an access video feature vector of the video accessed by the account within a predetermined time period, or an account attribute feature vector representing information of interest, age, occupation, and the like of the user corresponding to the account. The access video feature vector is obtained by learning the video sequence of the accessed video within the preset time, and the account attribute feature vector is obtained by learning the user sequence playing each video.

Specifically, when the account feature vector is an access video feature vector in the embodiment of the present disclosure, in order to facilitate management of videos in the video application, a video identifier, such as a video ID (Identity, Identity number), corresponding to each video is uploaded in the video application. After the server 102 obtains the video recommendation request, the video identifier of the video accessed by the video application within the predetermined time period in the near future may be obtained according to the video request, that is, the video identifier of the video accessed by the terminal device 101 in the video application within the predetermined time period corresponding to the account is obtained, where the predetermined time period may be 1 day, 3 days, a week, and the like. And then, according to the information of the account carried in the video recommendation request, determining the video identifier of the video accessed by the account within the preset time length, and thus determining the video feature vector of the video watched by the user within the preset time length, namely the accessed video feature vector, according to the preset mapping relationship between the video identifier and the video feature vector.

In step 202, a constructed video recommendation index is obtained, where the video recommendation index includes M layers, each layer includes a preset feature vector, an nth layer of the M layers includes a preset feature vector included in an (N-1) th layer, M is a natural number, and N is a natural number greater than 1 and smaller than M.

In the embodiment of the present disclosure, the video recommendation index is an index including multiple layers, for example, the video recommendation index may be a video recommendation index that is built step by step based on an HNSW algorithm and includes M layers, where M is greater than or equal to 2, an mth layer is a highest layer of the video recommendation index and includes all preset feature vectors corresponding to the video recommendation index, and an nth layer of the M layers includes preset feature vectors included in an (N-1) th layer, that is, in two adjacent layers, a preset feature vector of a higher layer of the two adjacent layers includes a preset feature vector of a lower layer.

Specifically, referring to fig. 3, the video recommendation index in the embodiment of the disclosure may have a stair-stepped structure as shown in fig. 3, the maximum level Lmax layer may also be referred to as the highest level, that is, the mth layer as described above, the Lmax layer includes all the predetermined eigenvectors corresponding to the video recommendation index, and the number of the predetermined eigenvectors included in the topmost L0 layer is the smallest, wherein the number of the predetermined eigenvectors included in the Lmax layer to the L0 layer decreases from layer to layer according to a predetermined increasing rate, and the predetermined eigenvector included in the Lmax-1 layer is included in the Lmax layer, that is, the predetermined eigenvector included in the low layer also appears in the high layer adjacent to the low layer. That is, the preset feature vector included in the Lmax-1 level may be selected from the Lmax level according to a predetermined rate-increase ratio, and the sample video feature vector included in the Lmax-2 level may be selected from the Lmax-1 level according to the predetermined rate-increase ratio, and so on until the number of the preset feature vectors included in the L0 level is determined. The preset increasing rate can be any value greater than 0 and less than 1.

In the embodiment of the present disclosure, all accounts of a video application may use the same video recommendation index, or all accounts of the video application may be classified, and accounts of different categories correspond to different video recommendation indexes, for example, a new user and an old user of the video application correspond to respective video recommendation indexes, and a user uploading a video in the video application and a user never uploading a video correspond to different video recommendation indexes. The video recommendation indexes are constructed according to different video recommendation indexes corresponding to different users, so that the video recommendation indexes are more suitable for the user requirements, and when videos with recommendations are searched by using the video recommendation indexes, videos to be recommended can be acquired quickly, so that the user experience is improved.

Further, when the server 102 receives the video recommendation request, the video recommendation index corresponding to the account may be determined according to the corresponding account in the terminal device 101.

In this embodiment of the disclosure, the preset feature vector may be a preset video embedding feature vector or a preset account attribute embedding feature vector, where if the preset feature vector is the preset video feature vector, a part or all of videos of the video application stored in the server 102 may be input into a pre-established feature learning model for learning, so as to obtain the embedding feature vector of each video in the video application. The feature learning model may be a Deep Neural Network (DNN) model, a Skip-Gram model, or other models, and parameters of the training model may be set according to a requirement recommended by a video when the feature learning model is trained, for example, a click rate of the video (i.e., a ratio of a number of times a certain video is clicked to a number of times the certain video is displayed), a click rate (i.e., a ratio of a number of times the certain video is clicked to a number of times the certain video is viewed), a report rate (i.e., a ratio of a number of times the video is reported to a number of times the certain video is viewed), and the like, so as to obtain the feature learning model meeting the requirement of the user.

Furthermore, the server 102 may establish a mapping relationship between the video identifier and the video feature vector in a video application according to the obtained embedding feature vector. When the video feature vector changes along with the change of the parameters set by the feature training model, the mapping relation between the video identifier and the video feature vector changes, so that the video feature vector determined according to the video identifier is more accurate.

Further, if the preset feature vector is an account attribute feature vector, the account information and the browsing information of some or all accounts of the video application may be input into the pre-established feature learning model for learning, so as to obtain the embedding feature vectors of multiple accounts in the video application.

As an optional implementation manner, in the embodiment of the present disclosure, the video recommendation index may be performed according to the following steps:

in the first step, a preset feature vector set is obtained.

As described above, the preset feature vector may be a preset video embedding feature vector or a preset account attribute embedding feature vector, and then, the obtained preset feature vector set may be a set composed of preset video feature vectors or a set composed of preset account attribute feature vectors. Correspondingly, the video recommendation index may be constructed based on a preset video feature vector, or may be constructed based on a preset account attribute feature vector.

In the embodiment of the disclosure, when the video recommendation index is constructed based on the preset video feature vector, the preset video required for constructing the video recommendation index may be acquired from the server corresponding to the video application, and then the acquired preset video is input into the feature learning model for learning, so as to obtain the preset video feature vector set corresponding to the preset video; after the video identifier of the preset video is obtained, the preset video feature vector set corresponding to the preset video can be determined from the video identifier and video feature vector mapping relationship according to the video identifier of the preset video.

Correspondingly, when the video recommendation index is constructed based on the account attribute feature vector, preset account information required for constructing the video recommendation index can be acquired from a server corresponding to the video application, and then the acquired account information is input into the feature learning model for learning, so that an account attribute feature vector set corresponding to a preset account is obtained; or after the preset account information is obtained, according to the account identifier of the preset account, determining a preset account attribute feature vector set corresponding to the preset account from the preset account identifier and account attribute feature vector mapping relationship.

In the embodiment of the present disclosure, since a new account or a new video may be added at any time by a video application, in order to improve the recall rate of video recommendation, the preset feature vectors in the preset feature vector set may be updated at predetermined time intervals, and a specific updating manner will be specifically described below, and will not be described herein again.

And secondly, constructing an index level mapping table based on the budget characteristic vector set, wherein the index level mapping table records the number of index layers, the number of characteristic vectors distributed in each layer and the information of the layer to which the characteristic vectors belong.

In the embodiment of the present disclosure, after the preset feature vector set is obtained, all preset feature vectors in the preset feature vector set may be used as a highest layer (i.e., an mth layer) in the video recommendation index, so that the number of layers included in the video recommendation index, and the number of preset feature vectors included in each layer, may be determined according to a predetermined increment rate based on the number of preset video feature vector sets in the preset feature vector set. Furthermore, a corresponding number of preset feature vectors can be selected from the preset feature vector set according to the number of the preset feature vectors distributed in each layer, and then the distribution positions of the preset feature vectors in each layer in the index hierarchical mapping table can be determined according to the similarity between the preset feature vectors included in each layer, so as to construct the index hierarchical mapping table.

The selection of the corresponding preset feature vector may be any selection from a preset feature vector set, or sequential selection according to the arrangement order of the preset feature vectors in the preset feature vector set. Therefore, the regularity that each preset feature vector in the preset feature vector set randomly appears in each level of the video recommendation index can be reduced, and the video recall rate is improved.

Specifically, in the embodiment of the present disclosure, in order to facilitate searching for the preset feature vector, each preset feature vector in the preset feature vector set may be provided with a corresponding identification number, and the identification number is only used to distinguish different preset feature vectors and does not represent an arrangement sequence of each preset feature vector in the preset feature vector set. Thus, the arbitrary selection may be to randomly select a corresponding number of preset feature vectors from a preset feature vector set according to the identifiers of the preset feature vectors, for example, assuming that the preset feature vector set includes 5000 preset feature vectors, the number of the preset feature vectors corresponding to the lowest layer in the index hierarchical mapping table is 100, the first 100 preset feature vectors with an even identifier mantissa may be selected from the preset feature vector set, or the preset feature vectors with identifiers from 1 to 100 may be selected, and so on.

Correspondingly, if the preset feature vectors of the number corresponding to each level are sequentially selected according to the arrangement sequence of the preset feature vectors in the preset feature vector set, a preset feature vector random array can be generated according to the obtained preset feature vectors in the preset feature vector set, and then the preset feature vectors corresponding to the number of the preset feature vectors distributed in each level in the index level mapping table can be sequentially determined according to the arrangement sequence of the preset feature vectors in the random array. Therefore, when the index level mapping table is constructed, the identifier of the preset feature vector corresponding to each index level, which is determined according to the arrangement sequence of each preset feature vector in the random array, can be recorded in the corresponding index level, so that the corresponding preset feature vector can be acquired from the preset feature vector set according to the preset feature vector identifier.

For example, assuming that the preset feature vector set includes 1000 preset feature vectors, a random array with the preset feature vector quantity of 1000 may be generated, and the predetermined increment rate is 0.2, it may be determined that the index hierarchy mapping table includes 4 layers, and the quantity of the preset feature vectors included in each layer from the 1 st layer to the 4 th layer is respectively: 8. 40, 200 and 1000. Then, the 1000 preset feature vectors may be determined as the level 4 of the index level mapping table, and the identities of the 1000 preset feature vectors may be recorded in the level 4 of the index level mapping table. Then, according to the 1000 preset feature vector arrangement order, 200 preset feature vectors with the top arrangement order are determined as the preset feature vectors in the layer 3, and the identifiers of the 200 preset feature vectors are recorded as the layer 3 of the index hierarchy mapping table. Furthermore, 40 preset feature vectors in the 200 preset feature vectors that are arranged in the top order may be determined as the preset feature vectors in the layer 2, and the identifiers of the 40 preset feature vectors may be recorded as the layer 2 of the index hierarchy mapping table. Then, 8 preset feature vectors in the 40 preset feature vectors that are arranged in the front order may be determined as the preset feature vectors in the layer 1, and the identifiers of the 8 preset feature vectors are recorded as the layer 1 of the index hierarchy mapping table.

It should be noted that, in the embodiment of the present disclosure, the predetermined increment rate is in positive correlation with the number of levels of the index level mapping table, and the larger the predetermined increment rate is, the more the number of levels of the index level mapping table is, and the larger the number of levels of the video recommendation index is, the larger the memory occupied by the video recommendation index is, the longer the time required for constructing the video recommendation index is, so that in order to ensure the efficiency of video recommendation, when the predetermined increment rate is set, the specific setting may be performed according to actual requirements.

And thirdly, distributing the preset feature vectors into each layer according to the index level mapping table to obtain the video recommendation index.

In the embodiment of the disclosure, after the index hierarchy mapping table is constructed, the number of layers of the video recommendation index, the number of preset feature vectors distributed in each layer, and the preset feature vectors can be determined according to the index hierarchy mapping table, so that the preset feature vectors included in each layer of the video recommendation index can be arranged according to the similarity height relationship. Specifically, the similarity of the preset feature vectors can be represented by the distance between the preset feature vectors, and the higher the similarity is, the closer the arrangement distance of the preset feature vectors is, so as to obtain the video recommendation index.

Further, for convenience of understanding, the position of each preset video feature vector in the layer may be referred to as a node, how many preset feature vectors in the layer correspond to a plurality of nodes, and nodes in the hierarchy of other preset feature vectors that are close to one preset feature vector may be referred to as friends of the corresponding node of the preset feature vector, and the corresponding nodes of the preset feature vectors are linked with friends of the corresponding node. Because the more friend points, the higher the recommendation quality, but the lower the search efficiency, the number of friend points of the node corresponding to each preset feature vector can be set according to the requirement, and in order to give consideration to the quality and recommendation speed of video recommendation, the number of friend points corresponding to each preset feature vector can be set in a proper range.

For example, assuming that a preset video feature vector a is referred to as a node a, and a preset feature vector adjacent to the preset feature vector a includes preset feature vectors B-F, the nodes B-F may be referred to as friend points of the node a, and further, if the number of friend points of the node a is set to be 3, 3 nodes adjacent to the node a may be determined from the nodes B-F to be friend points of the node a.

In step 203, similarity comparison is performed between the account feature vector and a preset feature vector included in the video recommendation index layer by layer to obtain a comparison result, where an input of a first layer of the video recommendation index is the account feature vector, and an input of an nth layer is the comparison result of an (N-1) th layer.

In the embodiment of the present disclosure, after the server 102 determines a video recommendation index used for video recommendation, the obtained account feature vector may be compared with a preset feature vector in a first layer (i.e., a highest layer) of the video recommendation index, and then a first feature vector whose similarity to the account feature vector meets a preset condition is determined in the first layer of the video recommendation index and is output as a comparison result of the first layer.

Next, the comparison result of the first layer (i.e., the first feature vector) may be used as an input of a second layer in the video recommendation index, the first feature vector may be compared with preset feature vectors distributed in the second layer, the preset video feature vectors in the second layer whose similarity to the first feature vector meets a preset condition may be output as a comparison result of the second layer, and then the comparison result of the second layer may be used as an input of a third layer. That is to say, the account feature vector may be used as an input of a first layer of the video recommendation index, and M layers included in the video recommendation index are sequentially traversed according to a principle that a low-layer comparison result in two adjacent layers is a high-layer comparison object, so as to determine a comparison result in the M-th layer.

In the embodiment of the present disclosure, the similarity satisfying the preset condition may refer to that the similarity between the feature vector input into each layer and the feature vectors distributed in each layer is greater than a first predetermined threshold, for example, a preset feature vector having a similarity greater than 80% in each layer may be output as the comparison result of the layer. Or, after the feature vectors input into each layer and the feature vectors distributed in each layer are sorted according to the similarity, at least one preset feature vector of K top-sorted layers may be output as the comparison result of the layer. For example, 3 preset feature vectors ranked in the top 3 may be output as the comparison result, or the 1 st preset feature vector may be output as the comparison result, that is, the preset feature vector with the highest similarity is the comparison result.

When the preset condition is that the similarity ranks at the top K bits, determining the comparison result of each layer of the video recommendation index comprises comparing the distance between the feature vector input by each layer and the preset feature vectors distributed in the layer, determining the preset feature vector closest to the input feature vector as the preset feature vector with the highest similarity to the input feature vector, and taking the preset feature vector with the highest similarity as the comparison result of each layer. That is, the similarity between the preset feature vectors may be determined by comparing the distances between the preset feature vectors in each layer.

For example, assuming that the video recommendation index includes four layers, namely a first layer to a fourth layer, the first layer is the lowest, the number of the included preset feature vectors is the lowest, the fourth layer is the highest, the preset feature vectors include all the preset feature vectors in the preset feature vector set, and only one account feature vector is assumed, namely, the account feature vector a, then the preset feature vector B closest to the account feature vector a may be determined in the first layer as the preset feature vector with the highest similarity to the account feature vector a, then the preset feature vector C closest to the preset feature vector B may be determined in the second layer as the preset feature vector with the highest similarity to the preset feature vector B, then the preset feature vector D closest to the preset feature vector C may be determined in the third layer as the preset feature vector with the highest similarity to the preset feature vector B, and finally, determining a preset feature vector E closest to the preset feature vector D in the fourth layer as the preset feature vector with the highest similarity to the preset feature vector D, wherein the preset feature vector E is a comparison result output by the fourth layer.

In the embodiment of the present disclosure, the comparison result obtained by each layer may be one or multiple, according to the preset condition of similarity for obtaining the comparison result, and in two adjacent layers, when the comparison result of the lower layer comprises a plurality of preset feature vectors, after the comparison result of the lower layer is input into the adjacent upper layer and compared with the preset feature vectors in the upper layer in the similarity, two or more predetermined feature vectors that may appear in the lower layer comparison result have a common predetermined feature vector with a similarity satisfying a predetermined condition in the adjacent higher layer, and thus, when the comparison result in the high layer is determined, the preset feature vectors in which the repeatedly determined similarity meets the preset condition can be deleted, only one feature vector is reserved, therefore, the phenomenon that repeated preset characteristic vectors appear in the comparison result to influence the video retrieval efficiency is avoided.

In step 204, a video is recommended to the account according to the comparison result output from the mth layer.

In the embodiment of the present disclosure, the number of videos recommended to the account by the video application at a time may be set, for example, 20, 50, and the like. When videos are recommended to the account after the comparison result output by the Mth layer is obtained, if the number of videos corresponding to the preset feature vector determined according to the comparison result is equal to the preset number of videos recommended to the account, the videos corresponding to the preset video feature vector can be directly recommended to the user as target videos; if the number of videos corresponding to the preset feature vector determined according to the comparison result is larger than the preset number of videos recommended to the account, the preset feature vector with the similarity degree meeting the preset condition with the account feature vector can be determined from the comparison result output by the Mth layer, and then the video corresponding to the determined preset feature vector is recommended to the account as the target video; if the number of videos corresponding to the preset feature vector determined according to the comparison result is smaller than the preset number of videos recommended to the account, after the comparison result is output on the mth layer, a preset feature vector (hereinafter referred to as a first preset feature vector) with the similarity meeting the preset condition with the comparison result can be determined in the mth layer, and then the videos corresponding to the preset video feature vector and the videos corresponding to the first preset feature vector in the comparison result output on the mth layer are recommended to the account as target videos. The preset conditions comprise at least one condition that the similarity is larger than a first preset threshold value, the similarity is ranked at the top K bits, and K is a natural number larger than 1.

Moreover, when account feature vectors used for retrieval are different, the video recommendation indexes used are also different, and further, the method of recommending videos to the account based on the price comparison result output by the Mth layer is also different. Therefore, the recall rate of the video can be improved, and the use experience of a user can be improved while the video recommendation efficiency is improved.

For example, when the account feature vector is an access video feature vector, the corresponding video recommendation index is a video recommendation index constructed based on preset video feature vectors, the comparison result output by the mth layer is a plurality of preset video feature vectors, and then a video corresponding to the preset video feature vectors can be recommended to the account as a target video, or when the number of videos (hereinafter referred to as a first video) corresponding to the comparison result (i.e., the plurality of preset video feature vectors) determined by the mth layer does not meet the recommendation number requirement, the preset video feature vectors having similarity meeting the preset condition with the comparison result can be determined in the mth layer, and then a corresponding number of videos (hereinafter referred to as a second video) can be determined from the videos corresponding to the preset video feature vectors in any manner (e.g., randomly selected and ranked in similarity), and recommending the first video and the second video as target videos to the account. Therefore, videos corresponding to the preset video feature vectors determined in the Mth layer can be directly recommended to the account, so that the video retrieval efficiency can be improved, and the video recommendation speed is higher.

Further, when the account feature vector is an account attribute feature vector, the corresponding video recommendation index is a video recommendation index constructed based on the preset account attribute feature vector, the comparison result output in the mth layer is a plurality of preset account attribute feature vectors, correspondingly, when a video is recommended to the account according to the comparison result output in the mth layer, a plurality of preset accounts corresponding to the plurality of preset account attribute feature vectors can be determined in advance, and then the video accessed by the plurality of preset accounts is recommended to the account as a target video.

Similarly, when the number of videos (hereinafter referred to as a third video) accessed according to a plurality of preset accounts does not meet the recommended number requirement, a plurality of accounts whose similarity to the corresponding account meets the preset condition in comparison and combination can be determined in the M-th layer, and then a corresponding number of videos (hereinafter referred to as a fourth video) are determined from the videos accessed by the accounts according to the above arbitrary manner, and the third video and the fourth video are recommended to the accounts as target videos. Because the account with higher account attribute feature vector similarity has higher information similarity, such as the same age, gender, and hobbies, the video recommendation performed according to the videos accessed by the similar accounts can better meet the user requirements and improve the user experience.

As an optional implementation manner, in the embodiment of the present disclosure, since a video application may add a new account over time, the account of the video application may upload a new video at any time, and also may delete some videos in a video library corresponding to the video application according to the feedback of the account and the uploading time of the video, the account number, the account preference, and the video number corresponding to the video application in the video application may change over time, the video feature vector corresponding to the video, and the account attribute feature vector corresponding to the account may also change with parameters set by the feature learning model, if after the video number and the account number corresponding to the video application change, video retrieval recommendation is still performed with a video recommendation index established before the change, a recall rate of video retrieval may also decrease, so that accuracy of videos recommended to a user is not high, and the recommended videos may be outdated old videos and cannot be recommended to the user by the newly uploaded videos, so that the user experience is reduced.

Therefore, in the embodiment of the present disclosure, the obtained preset feature vector set may be updated according to a predetermined time interval, and then the step of constructing the index level mapping table is returned, that is, the index level mapping table may be reconstructed based on the updated preset feature vector set, so as to improve the recall rate of video retrieval, the accuracy of video recommendation, and the experience of the user.

In the implementation of the present disclosure, the manner of updating the combination of the preset feature vectors includes deleting the preset feature vectors that do not change after a predetermined time period in the preset feature vector set; and/or adding a new preset feature vector to the preset feature vector set. Specifically, preset feature vectors of the video recommendation index can be scanned and constructed according to a preset time interval, and when the preset feature vectors which do not change after a preset time length exists in the scanned preset feature vectors, the preset feature vectors can be deleted; when a newly uploaded video or a newly added user exists in the video application, the feature vector corresponding to the newly added video or the user can be used as a preset feature vector to be added into a preset feature vector set; or adding a newly added preset feature vector when the preset feature vector is not changed after the preset time is deleted. Therefore, an updated preset feature vector set is obtained, so that the index level mapping table is reconstructed based on the updated preset feature vector set, and a new video recommendation index can be constructed according to the reconstructed index level mapping table, so that videos recommended by the video recommendation index better meet the user requirements.

In the embodiment of the present disclosure, in order to make the preset feature vectors be able to appear at each level of the preset recommendation index at an opportunity so as to reduce the influence of the appearance sequence of the preset feature vectors on the recall rate and accuracy of video retrieval, before reconstructing the index level mapping table, the current arrangement sequence of each preset feature vector in the updated preset feature vector set may be randomly scattered, and then the video recommendation index is reconstructed based on the preset feature vectors after the scattering sequence according to the aforementioned step of constructing the index level mapping table.

That is to say, in the video recommendation process, the index level mapping table is reconstructed at regular time along with the updating of the preset feature vector, and then a new video recommendation index is constructed based on the reconstructed index level mapping table, so that the problem that the native HNSW does not support feature updating and deleting can be solved, the recall rate of the video is improved, the real-time performance of the recommended video can be ensured, and the experience of a user is improved.

As an optional implementation manner, in this embodiment of the disclosure, after determining a target video that needs to be recommended to an account according to the comparison result output at the mth layer of the video recommendation index, before performing video recommendation, the determined target video that is recommended to the account may be further processed according to one or a combination of the following items:

firstly, deleting repeated videos in a target video;

in the embodiment of the present disclosure, videos with different video identifiers and the same video content may exist in a video application, for example, videos that are repeatedly uploaded by a user or videos that are shot by multiple users based on the same scene at the same angle, because the video feature vectors obtained by training the videos with the same content may be the same or similar, when the videos are retrieved based on the video recommendation index described above, it is determined that videos with the same content may exist in a target video feature, and if the videos with the repeated content are recommended to an account, the use experience of the user may be reduced, so before video recommendation, the videos with the repeated content in the target video feature may be deleted, so that the recommended videos better meet the user requirements, and the user experience is improved.

Second, deleting the video with the video definition lower than a second preset threshold value in the target video.

Specifically, different users have different shooting levels and different pixels of shooting equipment, so that the quality of videos uploaded to video applications is different, and in order to ensure the quality of videos recommended to the users, videos with the definition lower than a preset threshold value in target videos can be deleted, so that videos with higher definition in the target videos are recommended to accounts, and the use experience of the users is improved.

Thirdly, the target video is compressed.

In this embodiment of the disclosure, the target video may be one video or may include a plurality of videos, and when the number of the target videos is large, in order to reduce a memory occupied by the target video and improve efficiency of sending the determined target video to the terminal device 101 by the server 102 shown in fig. 1, the target video may be compressed, so that the compressed video is sent to the terminal device 101, and the terminal device 101 receives the video, decompresses the video, and then displays the decompressed video in a recommendation list of a video application, so that an account can view the video.

Therefore, according to the method, the video recommendation method and the device can utilize the constructed layered video recommendation index to perform video recommendation, after a video recommendation request is received, the corresponding account feature vector can be obtained according to the video recommendation request, wherein the video recommendation request carries information of an account of a video to be recommended, the account feature vector corresponds to the account information of the video to be recommended, the obtained account feature vector can be input into the video recommendation index, similarity comparison is performed between the account feature vector and the preset feature vector included in the video recommendation index layer by layer according to an index mode of a comparison result of a high-level multiplexing adjacent low-level, a comparison result of which the similarity meets a preset condition is obtained, and the video is recommended to the account according to the comparison result output by the highest level. Therefore, compared with the conventional tree-based video recommendation index, the video recommendation index has the advantages that the recall rate of video retrieval can be improved, and the quality of video recommendation service is improved. In addition, since the video recommendation index in the embodiment of the disclosure can be used for reconstructing a new video recommendation index by using the updated preset feature vector after the preset feature vector is updated at regular time, the problem that the original HNSW does not support feature vector updating and deleting can be solved, and then the recall rate of video retrieval can be further improved, so that the recommended video better meets the user requirements, and the use experience of the user is further improved.

Fig. 4a is a block diagram illustrating a video recommendation apparatus according to an exemplary embodiment, referring to fig. 4a, the apparatus including: receiving unit 401, obtaining unit 402, determining unit 403 and recommending unit 404, wherein:

the receiving unit 401 is configured to execute acquiring a corresponding account feature vector according to a received video recommendation request, where the video recommendation request carries information of an account of a video to be recommended, and the account feature vector corresponds to the information of the account;

an obtaining unit 402 configured to perform obtaining of a constructed video recommendation index, where the video recommendation index includes M layers, each layer includes a preset feature vector, an nth layer of the M layers includes a preset feature vector included in an (N-1) th layer, M is a natural number, and N is a natural number greater than 1 and smaller than M;

a determining unit 403, configured to perform similarity comparison between the account feature vector as an input and a preset feature vector included in the video recommendation index layer by layer, so as to obtain a comparison result, where the input for a first layer of the video recommendation index is the account feature vector, and the input for an nth layer is the comparison result for an (N-1) th layer;

a recommending unit 404 configured to perform recommending a video to the account according to the comparison result output from the mth layer.

In a possible implementation manner, please refer to fig. 4b, the video recommendation apparatus in fig. 4b further includes a construction unit 405, the video recommendation index is constructed by the construction unit 405 in the following manner, and the construction unit 405 is configured to perform: acquiring a preset characteristic vector set; constructing an index level mapping table based on a preset feature vector set, wherein the index level mapping table records the number of index layers, the number of feature vectors distributed in each layer and information of the layer to which the feature vectors belong; and distributing the preset feature vectors to each layer according to the index level mapping table to obtain the video recommendation index.

In a possible implementation, the video recommendation apparatus shown in fig. 4b further includes an updating unit 406, where the updating unit 406 is configured to perform: updating a preset characteristic vector set at preset time intervals; and updating the index level mapping table based on the updated preset feature vector set.

In a possible implementation, the updating unit 406 shown in fig. 4b is specifically configured to perform: deleting the preset feature vectors which are not changed after a preset time length in the preset feature vector set; and/or adding a new preset feature vector to the preset feature vector set.

In one possible implementation, the building unit 405 shown in fig. 4b is specifically configured to perform: determining the number of layers included in the index level mapping table and the number of preset feature vectors distributed in each layer according to a preset increment rate; selecting a corresponding number of preset feature vectors from a preset feature vector set according to the number of the preset feature vectors distributed in each layer; and determining the distribution position of the preset feature vectors in each layer in the index hierarchical mapping table according to the similarity height relation among the preset feature vectors included in each layer.

In a possible implementation, the building unit 405 is further configured to perform: and randomly selecting a corresponding number of preset eigenvectors from the preset eigenvector set, or sequentially selecting a corresponding number of preset eigenvectors according to the arrangement sequence of each preset eigenvector in the preset eigenvector set.

In a possible implementation, the video recommendation apparatus shown in fig. 4b further includes a sorting unit 407, where the sorting unit 407 is configured to perform: when the corresponding number of preset characteristic vectors are sequentially selected according to the arrangement sequence of each preset characteristic vector in the preset characteristic vector set, the current arrangement sequence of each preset characteristic vector in the updated preset characteristic vector set is randomly scattered.

In a possible implementation, the recommending unit 404 is further configured to perform: determining a preset feature vector of which the similarity with the account feature vector meets a preset condition from a comparison result output by the Mth layer; and recommending the video corresponding to the determined preset characteristic vector to the account as a target video.

In one possible embodiment, the preset condition includes one of the following conditions: the similarity is greater than a first predetermined threshold; the similarity is ordered at the top K bits, and K is a natural number greater than 1.

In a possible implementation, the determining unit 403 is specifically configured to perform: when the preset condition is that the similarity is ranked at the top K, comparing the distance between the feature vector input by each layer and the preset feature vectors distributed in the layer; and determining the preset feature vector closest to the input feature vector as the preset feature vector with the highest similarity to the input feature vector, and taking the preset feature vector with the highest similarity as the comparison result of each layer.

In a possible implementation, the video recommendation apparatus shown in fig. 4b further includes a processing unit 408, where the processing unit 408 is configured to process the target video recommended to the account according to one or a combination of the following conditions:

deleting the video with repeated content in the target video; or,

and compressing the target video.

With regard to the video recommendation apparatus in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

The division of the modules in the embodiments of the present disclosure is illustrative, and is only a logical function division, and there may be another division manner in actual implementation, and in addition, each functional module in each embodiment of the present disclosure may be integrated in one processor, may also exist alone physically, or may also be integrated in one module by two or more modules. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode.

Fig. 5 is a schematic diagram illustrating a structure of a video recommendation server, such as the server 102 shown in fig. 1, according to an exemplary embodiment. As shown in fig. 5, the video recommendation server in the embodiment of the present disclosure includes at least one processor 501, a memory 502 and a communication interface 503, where the memory 502 and the communication interface 503 are connected to the at least one processor 501, and a specific connection medium between the processor 501 and the memory 502 is not limited in the embodiment of the present disclosure, and in fig. 5, the processor 501 and the memory 502 are connected by a bus 500 as an example, the bus 500 is represented by a thick line in fig. 5, and connection manners between other components are only schematically illustrated and are not limited. The bus 500 may be divided into an address bus, a data bus, a control bus, etc., and is shown with only one thick line in fig. 5 for ease of illustration, but does not represent only one bus or one type of bus.

In the embodiment of the present disclosure, the memory 502 stores instructions executable by the at least one processor 501, and the at least one processor 501 may execute the steps included in the aforementioned video recommendation method by executing the instructions stored in the memory 502.

The processor 501 is a control center of the video recommendation server, and may connect various parts of the entire video recommendation server by using various interfaces and lines, and perform various functions and process data of the computing device by running or executing instructions stored in the memory 502 and calling data stored in the memory 502, thereby performing overall monitoring on the computing device. Optionally, the processor 501 may include one or more processing units, and the processor 501 may integrate an application processor and a modem processor, wherein the processor 501 mainly processes an operating system, a user interface, an application program, and the like, and the modem processor mainly processes wireless communication. It will be appreciated that the modem processor described above may not be integrated into the processor 501. In some embodiments, processor 501 and memory 502 may be implemented on the same chip, or in some embodiments, they may be implemented separately on separate chips.

The processor 501 may be a general-purpose processor, such as a Central Processing Unit (CPU), digital signal processor, application specific integrated circuit, field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or the like, that may implement or perform the methods, steps, and logic blocks disclosed in embodiments of the present disclosure. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present disclosure may be embodied directly in a hardware processor, or in a combination of hardware and software modules.

Memory 502, which is a non-volatile computer-readable storage medium, may be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The Memory 502 may include at least one type of storage medium, and may include, for example, a flash Memory, a hard disk, a multimedia card, a card-type Memory, a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Programmable Read Only Memory (PROM), a Read Only Memory (ROM), a charge Erasable Programmable Read Only Memory (EEPROM), a magnetic Memory, a magnetic disk, an optical disk, and so on. The memory 502 is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory 502 in the disclosed embodiments may also be circuitry or any other device capable of performing a storage function to store program instructions and/or data.

The communication interface 503 is a transmission interface capable of performing communication, and may receive data or transmit data through the communication interface 503, for example, may receive a video recommendation request transmitted by a terminal device through an account through the communication interface 503, and may also transmit video retrieved according to a video recommendation index to other devices through the communication interface 503.

Referring to FIG. 6, a further block diagram of the video recommendation server is shown, which further includes a basic input/output system (I/O system) 601 to facilitate the transfer of information between the various devices within the video recommendation server, and a mass storage device 605 for storing an operating system 602, application programs 603, and other program modules 604.

The basic input/output system 601 comprises a display 606 for displaying information and an input device 607, such as a mouse, keyboard, etc., for a user to input information. Wherein a display 606 and an input device 607 are connected to the processor 501 via a basic input/output system 601 connected to the system bus 500. The basic input/output system 601 may also include an input/output controller for receiving and processing input from a number of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, an input-output controller may also provide output to a display screen, a printer, or other type of output device.

The mass storage device 605 is connected to the processor 501 through a mass storage controller (not shown) connected to the system bus 500. The mass storage device 605 and its associated computer-readable media provide non-volatile storage for the server package. That is, the mass storage device 605 may include a computer-readable medium (not shown), such as a hard disk or CD-ROM drive.

According to various embodiments of the present disclosure, the video recommendation server package may also be run by a remote computer connected to a network through a network such as the internet. That is, the computing device may be connected to the network 608 via the communication interface 503 coupled to the system bus 500, or may be connected to another type of network or remote computer system (not shown) using the communication interface 503.

Based on the foregoing embodiments, the present disclosure provides a storage medium, and when an instruction in the storage medium is executed by a processor of a video recommendation server, the video recommendation server is enabled to execute a video recommendation method that implements any one of the video recommendation methods described above in the present disclosure or any one of the video recommendation methods that may be involved in any one of the video recommendation methods.

In some possible implementations, various aspects of the video recommendation method provided by the embodiments of the present disclosure may also be implemented in a form of a program product including program code for causing a computer to perform the steps of the video recommendation method according to various exemplary implementations of the present disclosure described above when the program product runs on the computer.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for video recommendation, comprising:

2. The method of claim 1, wherein the video recommendation index is constructed by:

acquiring a preset characteristic vector set;

3. The method of claim 2, wherein after obtaining a set of preset feature vectors, the method further comprises:

updating the preset characteristic vector set at preset time intervals;

4. The method of claim 3, wherein updating the set of preset feature vectors comprises:

and adding a new preset feature vector to the preset feature vector set.

5. The method of claim 3, wherein constructing an index hierarchy mapping table based on the set of predetermined feature vectors comprises:

6. The method of claim 5, wherein selecting a corresponding number of predetermined eigenvectors comprises:

7. The method according to claim 6, wherein if a corresponding number of preset eigenvectors are sequentially selected according to the arrangement order of each preset eigenvector in the preset eigenvector set, before reconstructing the video recommendation index after updating the preset eigenvector set according to the preset eigenvector set, the method further comprises:

8. A video recommendation apparatus, comprising:

9. A video recommendation server, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the video recommendation method of any of claims 1-7.

10. A storage medium, wherein instructions in the storage medium, when executed by a processor of a video recommendation server, enable the video recommendation server to perform the video recommendation method of any one of claims 1-7.