Method, device, equipment and storage medium for constructing video feature representation

Abstract

The present application discloses a method, apparatus, device and storage medium for constructing a video feature representation, and relates to the field of Internet technology. The method comprises: obtaining a first video and n second videos, where the second video is a video related to the first video, and n is a positive integer; obtaining feature representations of n second videos, where the feature representations of the second videos are used to represent the hidden features of the second videos; and constructing a feature representation of the first video based on the feature representations of the second videos. The present application constructs a feature representation of the first video by utilizing the feature representations of the second videos, and links the feature representations of the first video with the feature representations of the second videos to complete the construction of the feature representation of the first video; and uses the feature representations of multiple short videos related to the long video to construct the feature representation of the long video, thereby solving the problem of difficulty in extracting key information from the video during the construction of the feature representation of the long video.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method, apparatus, device, and storage medium for constructing a video feature representation. Background Art

[0002] Video feature representation can use mathematical vectors to record the feature information of the video, and use video feature representation to make video management faster.

[0003] In related technologies, the construction of short video feature representation usually involves extracting key content from the short video, using a small amount of key content to represent the information contained in the short video, and converting the key content into a feature vector to obtain the short video feature representation.

[0004] Compared with short videos, long videos contain richer and more diverse information. Key information in long videos is difficult to extract, and the feature representation ability constructed for long videos using the above method is poor. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for constructing a video feature representation. The technical solution is as follows:

[0006] According to one aspect of the present application, a method for constructing a video feature representation is provided, the method comprising:

[0007] Obtain a first video and n second videos, where the second videos are videos related to the first video, and n is a positive integer;

[0008] Obtaining feature representations of the n second videos, where the feature representations of the second videos are used to represent latent features of the second videos;

[0009] A feature representation of the first video is constructed based on the feature representation of the second video, where the feature representation of the first video is used to represent hidden features of the first video.

[0010] According to another aspect of the present application, a device for constructing a video feature representation is provided, the device comprising:

[0011] An acquisition module, configured to acquire a first video and n second videos, where the second videos are videos related to the first video, and n is a positive integer;

[0012] The acquisition module is further configured to acquire feature representations of the n second videos, where the feature representations of the second videos are used to represent hidden layer features of the second videos;

[0013] A construction module is used to construct a feature representation of the first video based on the feature representation of the second video, where the feature representation of the first video is used to represent the hidden layer features of the first video.

[0014] In an optional design of the present application, the building block is further used to:

[0015] A feature representation of the first video is constructed based on an average of the feature representations of the n second videos.

[0016] In an optional design of the present application, the building block is further used to:

[0017] Obtaining a feature weight of the second video, where the feature weight is used to indicate a degree of relevance between the second video and the first video;

[0018] A feature representation of the first video is constructed based on a weighted average of the feature representations of the n second videos, where the weighted average of the feature representations of the n second videos is determined based on the feature representation of the second video and a feature weight of the second video.

[0019] In an optional design of the present application, obtaining the feature weight of the trailer video includes:

[0020] Obtaining similarity information of the second video, wherein the similarity information is used to describe the degree of similarity between the second video and the first video; and determining a feature weight of the second video based on the similarity information of the second video;

[0021] and / or,

[0022] Obtain playback information of the second video, wherein the playback information is used to describe the playback status of the second video; and determine a feature weight of the second video based on the playback information of the second video.

[0023] In an optional design of the present application, the second video includes a trailer video and / or a recommended video, wherein the recommended video is a video recommended based on the video attributes of the first video.

[0024] In an optional design of the present application, the acquisition module is further configured to:

[0025] Obtaining the first video;

[0026] Obtain n trailer videos according to the first video.

[0027] In an optional design of the present application, the acquisition module includes:

[0028] an acquiring unit, configured to acquire a plurality of trailer videos according to the first video;

[0029] A screening unit is used to screen n trailer videos from the multiple trailer videos.

[0030] In an optional design of the present application, the screening unit is further used for:

[0031] From the plurality of trailer videos, excluding the trailer videos whose feature weights are lower than a first target threshold, to obtain n trailer videos;

[0032] The feature weight of the trailer video is used to represent the degree of relevance between the trailer video and the first video;

[0033] In an optional design of the present application, the acquisition module is further configured to:

[0034] Obtain the first video; and recommend n recommended videos based on the first video;

[0035] and / or,

[0036] Obtain the first video and the trailer video of the first video; and recommend n recommended videos based on the trailer video.

[0037] In an optional design of the present application, the screening unit is further used for:

[0038] Recommending multiple recommended videos, and selecting n recommended videos from the multiple recommended videos.

[0039] In an optional design of the present application, the screening unit is further used for:

[0040] excluding, from the plurality of recommended videos, the recommended videos whose feature weights are lower than a second target threshold, to obtain n recommended videos;

[0041] The feature weight of the recommended video is used to indicate the relevance between the recommended video and the first video.

[0042] In an optional design of the present application, the feature representation of the second video includes at least one of the following feature representations: a feature representation of an image dimension, a feature representation of an audio dimension, and a feature representation of a text dimension.

[0043] According to another aspect of the present application, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the method for constructing a video feature representation as described above.

[0044] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the method for constructing a video feature representation as described above.

[0045] According to another aspect of the present application, a computer program product or computer program is provided, wherein the computer program product or computer program includes computer instructions, wherein the computer instructions are stored in a computer-readable storage medium, and a processor reads and executes the computer instructions from the computer-readable storage medium to implement the method for constructing the video feature representation as described above.

[0046] The beneficial effects of the technical solution provided by this application include at least:

[0047] By utilizing the feature representation of the second video, the feature representation of the first video is constructed, and the related videos of the first video are fully utilized; the feature representation of the first video is linked with the feature representation of the second video to complete the construction of the feature representation of the first video; the problem of difficulty in extracting key information in the video during the construction of the feature representation of the long video is solved, and the feature representation of multiple short videos related to the long video is used to construct the feature representation of the long video. This not only utilizes the advantages of the short video feature representation construction method, but also reduces the complexity of constructing the feature representation of the long video. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0049] Figure 1 is a block diagram of a computer system used to construct a video feature representation provided by one embodiment of the present application;

[0050] Figure 2 is a schematic diagram of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0051] Figure 3 is a schematic diagram of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0052] Figure 4 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0053] Figure 5 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0054] Figure 6 This is an interface diagram of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0055] Figure 7 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0056] Figure 8 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0057] Figure 9 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0058] Figure 10 is a flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0059] Figure 11 is a structural block diagram of a device for constructing a video feature representation provided by an exemplary embodiment of the present application;

[0060] Figure 12 This is a structural block diagram of a server provided by an exemplary embodiment of the present application.

[0061] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0063] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0064] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0065] It should be understood that although the terms first, second, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, a first parameter may also be referred to as a second parameter, and similarly, a second parameter may also be referred to as a first parameter without departing from the scope of this disclosure. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0066] Figure 1 A schematic diagram of a computer system provided in one embodiment of the present application is shown. The computer system can be implemented as the system architecture of the method for constructing a video feature representation. The computer system may include: a terminal 100 and a server 200.

[0067] The terminal 100 can be an electronic device such as a mobile phone, tablet computer, vehicle-mounted terminal (vehicle computer), wearable device, smart home device, desktop computer, access control device, unmanned vending terminal, etc. The terminal 100 can be installed with a client that runs a target application. The target application can be a video application or other application that provides a construction function for video feature representation, which is not limited in this application. In addition, this application does not limit the form of the target application, including but not limited to App (Application, application) installed in the terminal 100, mini-program, etc., and can also be in the form of a web page.

[0068] The server 200 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The server 200 can be the background server of the target application, used to provide background services for the client of the target application; those skilled in the art will know that the number of the terminals 100 and / or servers 200 can be more or less. For example, there can be one terminal, or there can be dozens or hundreds of terminals, or a larger number. The embodiments of the present application do not limit the number and device types of the terminals 100 and / or servers 200.

[0069] In the method for constructing a video feature representation provided in the embodiment of the present application, the execution subject of each step may be a computer device, which refers to an electronic device with data calculation, processing and storage capabilities. Figure 1 Taking the implementation environment of the scheme shown as an example, the method for constructing the video feature representation can be executed by the terminal 100 (such as the client of the target application installed and running in the terminal 100 executes the method for constructing the video feature representation), or the method for constructing the video feature representation can be executed by the server 200, or the terminal 100 and the server 200 can interact and cooperate to execute it, and this application does not limit this.

[0070] Furthermore, the technical solution of this application can be combined with blockchain technology. For example, in the method for constructing a video feature representation disclosed in this application, some data involved (such as the first video, the second video, etc.) can be stored on a blockchain. The terminal 100 and the server 200 can communicate via a network, such as a wired or wireless network.

[0071] Figure 2 A schematic diagram of a method for constructing a video feature representation provided by an exemplary embodiment of the present application is shown. A long video 210 contains a lot of information, and the key information therein is difficult to extract, so it is impossible to directly obtain the feature representation 210a corresponding to the long video 210. The feature representation 210a corresponding to the long video 210 can be indirectly obtained through multiple short videos 220; the short video 220 includes: a first trailer video 222, a second trailer video 224, a first recommended video 226, and a second recommended video 228; wherein the short video 220 is a related video of the long video 210. Exemplarily, the short video 220 is related to at least one of the video attributes of the long video 210, and the video attributes include but are not limited to at least one of the following: video category, video theme, video release information, video time information, video content, and video status.

[0072] Compared to long video 210, multiple short videos 220 contain less information, making key information easier to extract and the feature representations corresponding to short videos 220 readily available. To separately describe the degree of relevance between each short video 220 and the long video, feature weights corresponding to each short video 220 are obtained. Specifically, first trailer video 222 corresponds to first feature representation 222a, and the feature weight of first trailer video 222 is first feature weight 222b; second trailer video 224 corresponds to second feature representation 224a, and the feature weight of second trailer video 224 is second feature weight 224b; first recommended video 226 corresponds to third feature representation 226a, and the feature weight of first recommended video 226 is third feature weight 226b; second recommended video 228 corresponds to fourth feature representation 228a, and the feature weight of second recommended video 228 is fourth feature weight 228b.

[0073] A weighted average operation is performed on the feature representation corresponding to the short video 220 and the corresponding feature weight to obtain the feature representation 210a corresponding to the long video 210.

[0074] Optional, such as Figure 3 As shown, the feature representation 210a corresponding to the long video 210 can be constructed multiple times based on the short video 220. First, a weighted average operation is performed on the first feature representation 222a, the first feature weight 222b, the second feature representation 224a, and the second feature weight 224b to obtain the feature representation 210a corresponding to the long video 210. Then, a weighted average operation is performed on the third feature representation 226a, the third feature weight 226b, the fourth feature representation 228a, and the fourth feature weight 228b to update the feature representation 210a corresponding to the long video 210.

[0075] Next, let’s introduce the video attributes:

[0076] Video attributes are inherent characteristics of a video. Exemplarily, video attributes include but are not limited to at least one of the following:

[0077] Video categories; for example: TV series, movies, variety shows, documentaries, short videos;

[0078] Exemplarily, the second video is related to the video category of the first video; for example, the first video and the second video are both TV series videos.

[0079] Video theme; for example, comedy, action, or spy;

[0080] Exemplarily, the second video is related to the video theme of the first video; for example, the first video is a comedy movie, the second video may be a comedy TV series, or the second video may be a documentary about the history of comedy.

[0081] Video distribution information; for example, video production information, video production information, video filing information, cast and crew information, video investment information, video collaboration information, and video uploader;

[0082] Exemplarily, the second video is related to the video distribution information of the first video; for example: the production unit of the first video is the same as the production unit of the second video; the producer of the first video holds shares in the collaborative unit of the second video; the second video is a short introductory video about the first video uploaded by the screenwriter of the first video.

[0083] Video time information: for example, the video's premiere time, video launch time, and video upload time;

[0084] Exemplarily, the second video is related to the video time information of the first video; for example, the premiere time of the first video and the upload time of the second video are in the same month.

[0085] Video content: For example, the time and location of the video, the language of the video;

[0086] Exemplarily, the second video is related to the video content of the first video; for example, the plots of the first video and the second video both take place in 1939; the plot of the first video takes place in the Shaanxi-Gansu-Ningxia region, and the plot of the second video takes place in the Yan'an region.

[0087] Video status: For example: video playback volume, video update status, video update frequency, video playback permissions, and video download permissions.

[0088] Exemplarily, the second video is related to the video status of the first video; for example, the first video is only allowed to be downloaded by member users, and the second video is only allowed to be viewed by member users.

[0089] Figure 4 A flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:

[0090] Step 310: Obtain a first video and n second videos;

[0091] The second video is a related video of the first video, and n is a positive integer. Optionally, the video length of the second video is shorter than the video length of the first video, or the video length of the second video is less than a first threshold, the video length of the first video is greater than a second threshold, and the first threshold is less than the second threshold.

[0092] Exemplarily, the second video is related to at least one of the video attributes of the first video; the video attributes include but are not limited to at least one of the following: video category, video theme, video release information, video time information, video content, and video status.

[0093] Exemplarily, the second video is usually an independent video, that is, the second video corresponds to a video file or a video file package; but it does not rule out the possibility that the second video is attached to the mother video, for example: the second video is attached to the first video, according to the indication information: the second video is the video content from the 15th second to the 23rd second in the first video.

[0094] It should be noted that the second video can be attached to the first video or to another parent video, and this embodiment does not impose any restrictions on this. The instruction information can be text information or compiled computer code information; this embodiment does not impose any restrictions on the form of the instruction information.

[0095] Exemplarily, when the second video is attached to the mother video, the indication information of the second video is determined based on the playback information of the second video; for example, the video portion of the mother video whose playback volume is higher than the quality threshold is used as the high-energy segment of the mother video, and the above high-energy segment is used as the second video.

[0096] Exemplarily, the first video and / or the second video can be a video or a video collection. For example, the first video is a TV series collection, and the TV series includes 20 episodes of TV series videos; the second video is a movie collection, and the movie includes three movie videos: Mandarin version, Cantonese version, and English version.

[0097] Optionally, the first video has more video information than the second video. The reason for the difference in the amount of video information between the first and second videos includes, but is not limited to, at least one of the following: video length, video clarity, video format, video bitrate, and video dimensions. Exemplarily, the second video is shorter than the first video, i.e., the first video is a long video and the second video is a short video. The first video is a virtual reality (VR) video and the second video is a two-dimensional video.

[0098] Step 320: Obtain feature representations of n second videos;

[0099] The feature representation of the second video is used to represent the hidden features of the second video. The hidden features of the second video are abstracted features of the second video and mapped to the dimensions of the hidden layer. The hidden features are used to distinguish different videos.

[0100] The feature representation of a video is typically a vector and / or matrix. Exemplarily, the method for obtaining the feature representation of the second video includes, but is not limited to, at least one of the following: encoding the second video, inputting the second video into a dimensional representation model, and inputting the second video into a graph embedding model. In this embodiment, the features of the second video can be mapped to multiple dimensions to obtain a feature representation of the second video; exemplary, the feature representation of the second video includes, but is not limited to, at least one of the following feature representations: a feature representation of an image dimension, a feature representation of an audio dimension, and a feature representation of a text dimension.

[0101] Step 330: constructing a feature representation of the first video based on the feature representation of the second video;

[0102] The feature representation of the first video is used to represent the hidden features of the first video. The hidden features of the first video are abstracted features of the first video and mapped onto the dimensions of the hidden layer. The hidden features are used to distinguish different videos. In this embodiment, the feature representation of the first video is constructed based on the feature representation of the second video, and the number of the second videos is one or more.

[0103] The feature representation of the first video includes the feature representation of the second video. The feature representation of the first video can directly include the feature representation of the second video, for example, by concatenating the feature representations of multiple second videos to obtain the feature representation of the first video. The feature representation of the first video can also indirectly include the feature representation of the second video, for example, by calculating the average value or superposition value of the feature representation of the second video. Taking the number of second videos as an example, the feature representation of the first video can be obtained by performing vector operations on the feature representations of multiple second videos. The feature representation of the first video can also be obtained by inputting the feature representation of the second video into a neural network, for example, into an artificial neural network (ANN).

[0104] Optionally, a feature representation of the first video is constructed based on an average of feature representations of n second videos. Illustratively, the average of the feature representations of the n second videos is calculated and set as the feature representation of the first video; wherein the dimension of the feature representation of the first video is generally the same as the dimension of the feature representation of the second video, but the dimension of the feature representations is not necessarily the same.

[0105] To sum up, the method provided in this embodiment constructs the feature representation of the first video by utilizing the feature representation of the second video, thereby making full use of the related videos of the first video; the feature representation of the first video is linked with the feature representation of the second video to complete the construction of the feature representation of the first video; it solves the problem of difficulty in extracting key information in the video during the construction of the feature representation of a long video, and uses the feature representations of multiple short videos related to the long video to construct the feature representation of the long video, which not only can utilize the advantages of the short video feature representation construction method, but also can reduce the complexity of constructing the feature representation of the long video.

[0106] Based on Figure 4 In an optional embodiment, the above step 330 can be implemented as the following two steps:

[0107] Obtain feature weights of the second video;

[0108] The feature weight is used to indicate the degree of relevance between the second video and the first video. Exemplarily, the feature weight of the second video is determined based on the similarity information and / or playback information of the second video. With the continuous development of Internet technology, new evaluation dimensions for the second video may emerge, and new information for determining the feature weight may emerge. This embodiment does not impose any restrictions on this.

[0109] Optionally, the range of the feature weight of the second video is: greater than or equal to 0 and less than or equal to 1.

[0110] Constructing a feature representation of the first video based on a weighted average of the feature representations of the n second videos;

[0111] The weighted average of the feature representations of the n second videos is determined based on the feature representation of the second video and the feature weight of the second video.

[0112] Exemplarily, the features of the first video are represented as follows:

[0113] lv=sum i (svp[i]×sv[i]);

[0114] Among them, lv represents the feature representation of the first video, sv represents the feature representation of the second video, svp represents the feature weight of the second video, i represents the i-th second video, and sum represents a summation operation.

[0115] To sum up, the method provided in this embodiment constructs the feature representation of the first video by utilizing the feature representation of the second video, and makes full use of the related videos of the first video; by introducing feature weights, the connection dimension between the feature representation of the first video and the feature representation of the second video is expanded, and the construction of the video feature representation is completed; it solves the problem of difficulty in extracting key information in the video, and improves the mechanism of the method for constructing video feature representation.

[0116] Based on Figure 4 In an optional embodiment, the above step 330 can be implemented as the following steps: Figure 5 As shown:

[0117] Step 330a: Obtain similarity information of the second video;

[0118] The similarity information of the second video is used to describe the degree of similarity between the second video and the first video; illustratively, for the dimensions describing the degree of similarity, please refer to the relevant content about video attributes above; optionally, the dimensions describing the degree of similarity can also include at least one of the following: the image content of the video, the audio content of the video, and the text content of the video.

[0119] Step 330b: Determine a feature weight of the second video based on the similarity information of the second video;

[0120] Exemplarily, the repetition rate between the second video and the first video is determined as the feature weight of the second video; optionally, by setting a target threshold, the content in the second video that exceeds the target threshold is determined to be content that repeats the first video.

[0121] When the dimension describing the degree of rereading is multiple dimensions, the repetition information of the multiple dimensions can be calculated to obtain the feature weight of the second video; or the repetition information of the multiple dimensions can be input into an artificial neural network (ANN) to obtain the feature weight of the second video.

[0122] Step 330c: Obtaining playback information of the second video;

[0123] The playback information of the second video is used to describe the playback status of the second video; optionally, the playback information of the second video is playback information associated with the first video; illustratively, the playback information of the second video is playback information of the first user watching the second video based on the first video after watching the first video. Exemplarily, watching the second video based on the first video can be achieved in at least one of the following ways: watching through push messages, watching on the recommendation page, watching in the recommendation area of ​​the user page, watching in a pop-up card; this embodiment does not impose any restrictive provisions on the method of watching the second video based on the first video. The above-mentioned method of watching the second video based on the first video can display the first video and the second video on the same user interface, or on different user interfaces; optionally, the first video and the second video can come from different applications and / or user terminals.

[0124] Step 330d: Determine a feature weight of the second video based on the playback information of the second video;

[0125] Exemplarily, based on the number of times the second video is played, the feature weight of the second video is determined as:

[0126] svp[i]=tp[i] / stp[n];

[0127] Among them, svp represents the feature weight of the second video, i represents the i-th second video, tp represents the number of times the second video is played, and stp[n] represents the total number of times n second videos are played.

[0128] Exemplarily, based on the number of times the second video has been played and the degree of completion of the play of the second video, the feature weight of the second video is determined as:

[0129] svp[i]=(tp[i]×cp[i]) / (stp[n]×acp[n]);

[0130] Among them, svp represents the feature weight of the second video, i represents the i-th second video, tp represents the number of times the second video is played, stp[n] represents the sum of the number of times the n second videos are played, cp represents the playback completion rate of the second video, and acp[n] represents the average value of the playback completion rate of the n second videos.

[0131] Optionally, based on the playback information of the first video and the playback information of the second video, the feature weight of the second video is determined; illustratively, based on the number of times the second video has been played, the playback completion degree of the second video, and the number of times the first video has been played, the feature weight of the second video is determined as:

[0132] svp[i]=(tp[i]×cp[i]) / ltp;

[0133] Among them, svp represents the feature weight of the second video, i represents the i-th second video, tp represents the number of times the second video is played, and ltp represents the number of times the first video is played.

[0134] Optionally, the feature weight of the second video is updated according to the effective playback threshold. For example, the feature weight of the second video is:

[0135] svp*[i]=svp[i]-Thre;

[0136] Among them, svp* represents the feature weight of the updated second video, and Thre represents the effective playback threshold; illustratively, the effective playback threshold can be set based on experience, or determined based on the playback information of the first video and / or the second video; with the expansion of the evaluation dimension of the effective playback threshold, the effective playback threshold can be set based on other information, and this embodiment does not make any restrictive provisions on this.

[0137] Optionally, when n second videos are divided into a first-category videos and b second-category videos, based on the number of times the second videos are played and the degree of completion of the play of the second videos, the feature weights of the second videos are determined as follows:

[0138] svp a =W1 / W;

[0139] svp b =W2 / W;

[0140] W1=sum a (tp a ×cp a );

[0141] W2=sum b (tp b ×cp b );

[0142] W=W1+W2;

[0143] Among them, svp a Represents the feature weight of a type of video, tp a Indicates the number of times a type of video is played, cp a Indicates the completion rate of a video. a Indicates the summation of a videos of the same category; svp b Represents the feature weight of the second-class video, tp b Indicates the number of times the second-category video is played, cp b Indicates the playback completion of the second category video, sum b Indicates the summation calculation of b second-class videos.

[0144] The first and second categories of videos are categorized based on the video attributes of the second video. By adjusting the granularity of the second video classification, the second video can be divided into more categories. For example, the first category of videos is a trailer video of the first video, and the second category of videos is a recommended video of the first video. The recommended video is a video recommended based on the video attributes of the first video.

[0145] It should be noted that the above exemplary description is only an example of the method for determining the feature weight of the second video. Those skilled in the art will understand that the feature weight of the second video can be determined by using at least one of the playback information such as comment information, sharing information, and praise information of the first video and / or the second video. This embodiment does not make any restrictive provisions on the evaluation dimension of the playback information.

[0146] It should be noted that, in this embodiment, the relationship between the first branch and the second branch is an and / or relationship. Only one of the first branch and the second branch can be executed, or both branches can be executed to form a new embodiment; wherein, the first branch includes step 334a and step 334b, and the second branch includes step 334c and step 334d. When both branches are executed, the feature weight of the second video is determined based on the similarity of the second video and the playback information of the second video; illustratively, the feature weight is obtained by calculating the similarity of the second video and the playback information of the second video, or by inputting the similarity of the second video and the playback information of the second video into the ANN; optionally, similar to when the two branches are executed separately, the value range of the feature weight is: greater than or equal to 0 and less than or equal to 1.

[0147] Step 330e: constructing a feature representation of the first video based on a weighted average of the feature representations of the n second videos;

[0148] The weighted average of the feature representations of the n second videos is determined based on the feature representations of the second videos and the feature weights of the second videos.

[0149] In summary, the method provided in this embodiment constructs a feature representation for the first video by utilizing the feature representation of the second video, fully utilizing the first video's related videos. It also determines feature weights based on similarity information, expanding the dimension of evaluating the connection between the first and second videos from the perspective of video content. It also determines feature weights based on playback information, expanding the dimension of evaluating the connection between the first and second videos from the perspective of video playback feedback. This completes the construction of a video feature representation, addresses the difficulty in extracting key information from videos, and improves the mechanism of constructing a video feature representation.

[0150] Next, let’s further introduce the second video:

[0151] Exemplarily, the second video includes but is not limited to a trailer video and / or a recommended video; the trailer video and the recommended video are respectively introduced below:

[0152] Trailer video;

[0153] The trailer video is obtained by editing and / or extracting the first video. The length of the trailer video is usually shorter than that of the first video, but it does not rule out the possibility that the length of the trailer video is longer than that of the first video.

[0154] Exemplarily, the editing operation includes but is not limited to at least one of the following:

[0155] Edit videos: cut, piece together, and add video clips;

[0156] Edit images: add special effects, add filters, change image size, change image orientation, change image clarity;

[0157] Edit audio: add tracks, delete tracks, replace tracks, change track properties;

[0158] Edit text: add subtitles, delete subtitles, replace subtitles, and change subtitle properties.

[0159] For example, the first video is the full video of the fourth episode of the TV series, and the trailer video is a video clip of the first character being knocked down in the fourth episode of the TV series; the first video is a TV series collection, and the trailer video is a trailer for the TV series collection.

[0160] Exemplarily, when a trailer video is attached to a parent video, the trailer video is obtained by extracting the first video, where the extraction operation is determined based on the indication information. Exemplarily, a portion of the parent video with a playback volume exceeding a quality threshold is used as a high-energy segment of the parent video, and the high-energy segment is used as the trailer video.

[0161] It should be noted that, in an embodiment of the present application, the trailer video may include content that does not exist in the first video, such as: shooting highlights of the first video, a trailer of the first video, and a promotional short video of the first video.

[0162] Recommended videos;

[0163] For example, the display of recommended videos can be achieved through at least one of the following methods: display in a push message, display on a recommendation page, display in the recommendation area of ​​a user page, or display in a pop-up card. This embodiment does not impose any restrictions on the display method of recommended videos. The above-mentioned display method of recommended videos can display the first video and the second video on the same user interface or on different user interfaces. Optionally, the first video and the second video can be from different applications and / or user terminals.

[0164] The recommended video is a video recommended based on the video attributes of the first video. For an introduction to the video attributes, please refer to the content of step 310 above.

[0165] For example, Figure 6 An interface diagram showing a method for constructing a video feature representation provided by an exemplary embodiment of the present application is shown; Figure 6 A method for displaying a first video, a trailer video, and a recommended video is shown. Figure 6 The interface diagram of a video player is shown. The video player is used to play a first video 410. A first recommended video 422, a second recommended video 424, and a third recommended video 426 of the first video 410 are displayed in a recommendation display area 420. A first trailer video 432, a second trailer video 434, and a third trailer video 436 of the first video 410 are displayed in a highlight content display area 430.

[0166] Next, obtaining the first video and n second videos is described in detail;

[0167] As mentioned in the previous section, the second video includes but is not limited to trailer videos and / or recommended videos. There are differences in how the trailer videos and recommended videos are obtained. This section will combine Figure 7 、 Figure 8 and Figure 9 The illustrated embodiment provides a detailed description of the process of obtaining the first video and n second videos, especially the process of obtaining the second video:

[0168] Trailer video;

[0169] Based on the above Figure 4 The embodiment shown, as Figure 7 As shown, step 310 includes the following two sub-steps:

[0170] Step 312: Obtain a first video;

[0171] Exemplarily, the first video may be obtained by directly obtaining a video file of the first video, or may be obtained indirectly, such as by obtaining a video link of the first video.

[0172] Step 314: Obtain n trailer videos based on the first video;

[0173] The trailer video is obtained by editing and / or extracting the first video. The length of the trailer video is usually shorter than that of the first video, but it does not rule out the possibility that the length of the trailer video is longer than that of the first video.

[0174] For example, the number of trailer videos can be set based on experience or determined based on the video attributes of the first video; this embodiment does not impose any restrictions on the number of trailer videos, that is, the value of n.

[0175] Optional, such as Figure 8 As shown, step 314 includes the following two sub-steps:

[0176] Step 314a: Obtain multiple trailer videos based on the first video;

[0177] Exemplarily, the multiple trailer videos may be all trailer videos of the first video, or may be a part of the trailer videos, wherein the number of the multiple trailer videos is greater than or equal to n.

[0178] Step 314b: Filtering n trailer videos from the plurality of trailer videos;

[0179] Exemplarily, screening of trailer videos may be performed randomly, or may be determined based on video attributes of the first video and / or the trailer video. Exemplarily, screening of trailer videos based on video attributes of the trailer video is used as an example for exemplary description:

[0180] For example, a method for screening trailer videos is as follows: from a plurality of trailer videos, exclude trailer videos whose feature weights are lower than a first target threshold, to obtain n trailer videos;

[0181] Among them, the feature weight of the trailer video is used to indicate the degree of relevance between the trailer video and the first video; for details on the method for determining the feature weight of the trailer video, please refer to the method for determining the feature weight of the second video above.

[0182] In this embodiment, the first target threshold can be set based on experience or determined based on the video attributes of the first video and / or the trailer video; the first target threshold is used to exclude trailer videos with low feature weights, and when constructing the feature representation of the first video based on the feature representation of the trailer video, the feature representation of the first video is constructed using the trailer video that has a strong correlation with the first video.

[0183] To sum up, the method provided in this embodiment constructs the feature representation of the first video by utilizing the feature representation of the trailer video, makes full use of the related videos of the first video, and improves the method of obtaining the related videos of the first video; links the feature representation of the first video with the feature representation of the trailer video to complete the construction of the video feature representation; solves the problem of difficulty in extracting key information from the video, and improves the mechanism of the method for constructing the video feature representation.

[0184] Recommended videos;

[0185] Based on the above Figure 4 The embodiment shown, as Figure 9 As shown, step 310 includes the following four sub-steps:

[0186] Step 316: Obtain the first video;

[0187] Exemplarily, the first video may be obtained by directly obtaining a video file of the first video, or may be obtained indirectly, such as by obtaining a video link of the first video.

[0188] Step 317: Recommend n recommended videos based on the first video;

[0189] The recommended video is a video that is recommended based on the video attributes of the first video.

[0190] For example, the number of recommended videos may be set based on experience or determined based on the video attributes of the first video. This embodiment does not impose any restrictions on the number of recommended videos, ie, the value of n.

[0191] Step 318: Obtain the first video and a trailer video of the first video;

[0192] Exemplarily, the first video can be obtained by directly obtaining the video file of the first video, or indirectly, such as by obtaining a video link of the first video. Similarly, the trailer video of the first video can be obtained directly or indirectly. The trailer video is obtained by editing and / or extracting the first video.

[0193] Step 319: Recommend n recommended videos based on the trailer video;

[0194] The recommended video is a video that is recommended based on the video attributes of the first video.

[0195] Exemplarily, the number of recommended videos may be set based on experience, or may be determined based on the video attributes of the first video and / or trailer video; this embodiment does not impose any restrictions on the number of recommended videos, ie, the value of n.

[0196] It should be noted that in this embodiment, the relationship between the first branch and the second branch is an and / or relationship. Either the first branch or the second branch can be executed, or both branches can be executed to form a new embodiment; wherein, the first branch includes steps 316 and 317, and the second branch includes steps 318 and 319. When both branches are executed, the number of recommended videos is n, and there is at least one recommended video recommended based on the first video, and there is at least one recommended video recommended based on the trailer video. The recommended video recommended based on the first video and the recommended video recommended based on the trailer video are usually different, but it does not rule out the possibility that the recommended videos are the same.

[0197] Optionally, recommending n recommended videos can be achieved by the following method:

[0198] Recommend multiple recommended videos and select n recommended videos from the multiple recommended videos;

[0199] Exemplarily, the multiple recommended videos may be all the recommended videos of the first video, or may be a part of the recommended videos, wherein the number of the multiple recommended videos is greater than or equal to n.

[0200] For example, the screening of recommended videos may be random screening, or may be determined based on a video attribute of at least one of the first video, the trailer video, and the recommended video. For example, screening recommended videos based on a video attribute of a recommended video is used as an example for explanation:

[0201] For example, a method for screening recommended videos is as follows: from multiple recommended videos, exclude recommended videos whose feature weights are lower than a second target threshold, to obtain n recommended videos;

[0202] The feature weight of the recommended video is used to indicate the relevance between the recommended video and the first video. For details on the method for determining the feature weight of the recommended video, please refer to the method for determining the feature weight of the second video described above.

[0203] In this embodiment, the second target threshold can be set based on experience or determined based on the video attributes of the first video and / or the recommended video; the second target threshold is used to exclude recommended videos with low feature weights, and when constructing the feature representation of the first video based on the feature representation of the recommended video, the feature representation of the first video is constructed using recommended videos that have a strong correlation with the first video.

[0204] To sum up, the method provided in this embodiment constructs the feature representation of the first video by utilizing the feature representation of the recommended video, makes full use of the related videos of the first video, and improves the method of obtaining the related videos of the first video; links the feature representation of the first video with the feature representation of the recommended video to complete the construction of the video feature representation; solves the problem of difficulty in extracting key information from the video, and improves the mechanism of the method for constructing the video feature representation.

[0205] Next, the feature representation of the second video is introduced in detail:

[0206] In the present application, the features of the second video can be mapped to multiple dimensions to obtain a feature representation of the second video; exemplarily, the feature representation of the second video includes but is not limited to at least one of the following feature representations: feature representation of image dimension, feature representation of audio dimension, and feature representation of text dimension.

[0207] Among them, the feature representation of the image dimension is to abstract the image features of the video and map them to the dimensions of the hidden layer to obtain the features; exemplarily, the image features include but are not limited to at least one of the following: image frames in the video, key frames in the video, image frames corresponding to the highlights, and image frames with the most playback times.

[0208] The feature representation of the audio dimension is to abstract the audio features of the video and map them to the dimensions of the hidden layer to obtain the features; for example, the audio features include but are not limited to at least one of the following: background music in the video, actor dialogue voice in the video, narration voice in the video, and the opening song of the video.

[0209] The feature representation of the text dimension is to abstract the text features of the video and map them to the dimensions of the hidden layer to obtain features; illustratively, the text features include but are not limited to at least one of the following: video title, video comments, and video introduction text.

[0210] The feature representation of the above dimensions is obtained by inputting the second video into the corresponding dimensional representation model. Taking the second video as an example, the feature representation of the second video is described in detail:

[0211] Exemplarily, the image feature representation of a short video is obtained by inputting the image frames of the short video into an artificial neural network; for example, the image frames are input into an efficient network model (Efficient Net) to realize the process of aggregating frame-level features for large-scale video classification.

[0212] Exemplarily, the audio feature representation of a short video is obtained by inputting the audio frame of the short video into an artificial neural network; for example, the audio frame is input into a VGGish speech feature extraction model to implement a process of aggregating frame-level features for large-scale video classification.

[0213] Exemplarily, the text feature representation of the short video is obtained by inputting the text content of the short video into an artificial neural network; for example, the text content is first tokenized, and the tokenized text content is input into the bidirectional encoder representation (Bidirectional Encoder Representations from Transformers, BERT) model of the transformer to obtain the text feature representation.

[0214] Optionally, when obtaining feature representations of the second video from multiple dimensions, the multiple feature representations are fused to obtain the feature representation of the second video, that is, the second video is described from multiple dimensions.

[0215] To sum up, the method provided in this embodiment constructs the feature representation of the first video by utilizing the feature representation of the second video, makes full use of the related videos of the first video, and improves the method of obtaining the feature representation of the second video; links the feature representation of the first video with the feature representation of the second video to complete the construction of the video feature representation; solves the problem of difficulty in extracting key information in the video, and improves the mechanism of the method for constructing the video feature representation.

[0216] Next, we will take the first video as a long video and the second video as a short video as an example to provide a detailed introduction:

[0217] Figure 10 A flowchart of a method for constructing a video feature representation provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:

[0218] Step 410: Obtain a long video;

[0219] Long videos are obtained on video platforms.

[0220] Step 420: Obtain multiple trailer videos based on the long video;

[0221] Exemplarily, trailer videos include but are not limited to: highlights of long videos, trailers of long videos, and short videos edited from long videos; trailer videos can be obtained from a video platform or from a short video library.

[0222] Step 430: Obtain multiple recommended videos;

[0223] Recommended videos are determined based on trailer videos and / or long videos. In this embodiment, recommended videos can be obtained multiple times. After the first recommended video is obtained, recommended videos can be determined based on at least one of the trailer videos, long videos, and recommended videos. For example, multiple recommended videos can be determined based on multiple recommended videos.

[0224] Step 440: Filter n short videos from the multiple trailer videos and the multiple recommended videos;

[0225] For example, the number of times trailer videos and recommended videos are played is used as the basis for screening n short videos. The feature weight of videos with high play counts is higher than that of videos with low play counts. For example, if the number of plays of the first recommended video exceeds the target threshold, the first recommended video is considered to be incapable of representing a long video and is not suitable as a recommended video for a long video. The first recommended video is excluded.

[0226] For example, in the cold start phase of a long video, for example, when the number of plays of the long video is less than the target threshold, the n short videos are all trailer videos of the long video; in the cold start phase of a long video, there is less user behavior, and only screening trailer videos as short videos is conducive to improving the accuracy of the feature representation of the long video.

[0227] During the normal startup phase of a long video, for example, when the number of plays of the long video is less than the target threshold, the n short videos include both trailer videos and recommended videos. Using both trailer videos and recommended videos as short videos is conducive to making the feature representation of the long video more complete.

[0228] Step 450: Obtain feature representations of n short videos and feature weights of n short videos;

[0229] The n short videos are input into a multi-dimensional representation model, and the multi-dimensional representation model is used to extract image feature representation, audio feature representation and text feature representation of the n short videos.

[0230] Step 460: constructing a feature representation of the long video based on the weighted average of the feature representations of the n short videos;

[0231] The weighted average of the feature representation of the short video is the product of the feature representation of the short video and the feature weight of the short video.

[0232] During the cold start phase of a long video, for example, when the number of plays of a long video is less than the target threshold, the n short videos are all trailer videos of the long video; that is, the feature representation of the long video is constructed based on the weighted average of the feature representations of the n trailer videos; during the normal startup phase of a long video, for example, when the number of plays of a long video is less than the target threshold, the n short videos include both trailer videos and recommended videos, that is, the feature representation of the long video is constructed based on the weighted average of the feature representations of the n short videos.

[0233] At different stages of a long video's launch, you can build and update the feature representation of the video multiple times. This feature representation can be used for long and / or short video recommendations, improving the effectiveness of video recommendations.

[0234] To sum up, the method provided in this embodiment constructs the feature representation of the long video by utilizing the feature representation of the short video, and makes full use of the related videos of the long video; the feature representation of the long video is linked with the feature representation of the short video to complete the construction of the feature representation of the long video; it solves the problem of difficulty in extracting key information in the video during the construction of the feature representation of the long video, and uses the feature representations of multiple short videos related to the long video to construct the feature representation of the long video, which not only can utilize the advantages of the short video feature representation construction method, but also can reduce the complexity of constructing the feature representation of the long video.

[0235] Those skilled in the art will appreciate that the above embodiments may be implemented independently, or the above embodiments may be freely combined to form new embodiments to implement the method for constructing video feature representation of the present application.

[0236] Figure 11 A block diagram of a device for constructing a video feature representation provided by an exemplary embodiment of the present application is shown. The device includes:

[0237] An acquisition module 510 is configured to acquire a first video and n second videos, where the second videos are videos related to the first video, and n is a positive integer;

[0238] The acquisition module 510 is further configured to acquire feature representations of the n second videos, where the feature representations of the second videos are used to represent latent features of the second videos;

[0239] The construction module 520 is used to construct the feature representation of the first video based on the feature representation of the second video, where the feature representation of the first video is used to represent the hidden layer features of the first video.

[0240] In an optional design of this embodiment, the building module 520 is further configured to:

[0241] A feature representation of the first video is constructed based on an average of the feature representations of the n second videos.

[0242] In an optional design of this embodiment, the building module 520 is further configured to:

[0243] Obtaining a feature weight of the second video, where the feature weight is used to indicate a degree of relevance between the second video and the first video;

[0244] A feature representation of the first video is constructed based on a weighted average of the feature representations of the n second videos, where the weighted average of the feature representations of the n second videos is determined based on the feature representation of the second video and a feature weight of the second video.

[0245] In an optional design of this embodiment, obtaining the feature weight of the trailer video includes:

[0246] Obtaining similarity information of the second video, wherein the similarity information is used to describe the degree of similarity between the second video and the first video; and determining a feature weight of the second video based on the similarity information of the second video;

[0247] and / or,

[0248] Obtain playback information of the second video, wherein the playback information is used to describe the playback status of the second video; and determine a feature weight of the second video based on the playback information of the second video.

[0249] In an optional design of this embodiment, the second video includes a trailer video and / or a recommended video, wherein the recommended video is a video recommended based on video attributes of the first video.

[0250] In an optional design of this embodiment, the acquisition module 510 is further configured to:

[0251] Obtaining the first video;

[0252] Obtain n trailer videos according to the first video.

[0253] In an optional design of this embodiment, the acquisition module 510 includes:

[0254] An acquiring unit 512 is configured to acquire a plurality of trailer videos according to the first video;

[0255] The screening unit 514 is used to screen n trailer videos from the multiple trailer videos.

[0256] In an optional design of this embodiment, the screening unit 514 is further configured to:

[0257] From the plurality of trailer videos, excluding the trailer videos whose feature weights are lower than a first target threshold, to obtain n trailer videos;

[0258] The feature weight of the trailer video is used to represent the degree of relevance between the trailer video and the first video;

[0259] In an optional design of this embodiment, the acquisition module 510 is further configured to:

[0260] Obtain the first video; and recommend n recommended videos based on the first video;

[0261] and / or,

[0262] Obtain the first video and the trailer video of the first video; and recommend n recommended videos based on the trailer video.

[0263] In an optional design of this embodiment, the screening unit 514 is further configured to:

[0264] Recommending multiple recommended videos, and selecting n recommended videos from the multiple recommended videos.

[0265] In an optional design of this embodiment, the screening unit 514 is further configured to:

[0266] excluding, from the plurality of recommended videos, the recommended videos whose feature weights are lower than a second target threshold, to obtain n recommended videos;

[0267] The feature weight of the recommended video is used to represent the relevance between the recommended video and the first video;

[0268] In an optional design of this embodiment, the feature representation of the second video includes at least one of the following feature representations: a feature representation of an image dimension, a feature representation of an audio dimension, and a feature representation of a text dimension.

[0269] It should be noted that the device provided in the above embodiment only uses the division of the above-mentioned functional modules as an example to implement its functions. In actual applications, the above-mentioned functions can be assigned to different functional modules according to actual needs, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0270] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method; the technical effects achieved by each module performing operations are the same as the technical effects in the embodiment of the method, and will not be elaborated here.

[0271] An embodiment of the present application also provides a computer device, which includes: a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the video feature representation construction method provided by the above-mentioned method embodiments.

[0272] The computer device may be a terminal or a server. For example, in this embodiment, the computer device is a server as an example for detailed description: Figure 12 This is a structural block diagram of a server provided by an exemplary embodiment of the present application.

[0273] Typically, the server 2300 includes a processor 2301 and a memory 2302 .

[0274] The processor 2301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 2301 may be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor 2301 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 2301 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 2301 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0275] The memory 2302 may include one or more computer-readable storage media, which may be non-transitory. The memory 2302 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 2302 is used to store at least one instruction, which is executed by the processor 2301 to implement the video data tracking method provided in the method embodiment of the present application.

[0276] In some embodiments, the server 2300 may further optionally include an input interface 2303 and an output interface 2304. The processor 2301, the memory 2302, and the input interface 2303 and the output interface 2304 may be connected via a bus or signal lines. Each peripheral device may be connected to the input interface 2303 and the output interface 2304 via a bus, a signal line, or a circuit board. The input interface 2303 and the output interface 2304 may be used to connect at least one peripheral device related to input / output (I / O) to the processor 2301 and the memory 2302. In some embodiments, the processor 2301, the memory 2302, and the input interface 2303 and the output interface 2304 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 may be implemented on a separate chip or circuit board, which is not limited in the embodiments of the present application.

[0277] Those skilled in the art will understand that the above structure does not constitute a limitation on the server 2300 , and may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0278] In an exemplary embodiment, a chip is further provided. The chip includes a programmable logic circuit and / or program instructions. When the chip is run on a computer device, it is used to implement the method for constructing the video feature representation described in the above aspects.

[0279] In an exemplary embodiment, a computer program product or computer program is also provided. The computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods for constructing video feature representations provided in the above-described method embodiments.

[0280] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one program code. When the program code is loaded and executed by a processor of a computer device, the method for constructing the video feature representation provided by the above-mentioned method embodiments is implemented.

[0281] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0282] Those skilled in the art will appreciate that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented using hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any media that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0283] The above description is merely an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for constructing a video feature representation, characterized in that: The method comprises: Obtain a first video and n second videos, where the video length of the second videos is shorter than that of the first video, and n is a plurality of second videos. The second videos include recommended videos, or the second videos include trailer videos of the first video and the recommended videos, and the recommended videos are videos recommended based on the video attributes of the first video. The first video is obtained, and n recommended videos are recommended based on the first video. Furthermore, the first video and the trailer videos of the first video are obtained, and n recommended videos are recommended based on the trailer videos. Obtaining feature representations of the n second videos, where the feature representations of the second videos are used to represent latent features of the second videos; A feature representation of the first video is constructed based on the feature representation of the second video, where the feature representation of the first video is used to represent hidden features of the first video.

2. The method according to claim 1, characterized in that The constructing the feature representation of the first video based on the feature representation of the second video includes: A feature representation of the first video is constructed based on an average of the feature representations of the n second videos.

3. The method according to claim 2, characterized in that The constructing the feature representation of the first video based on the average value of the feature representations of the n second videos includes: Obtaining a feature weight of the second video, where the feature weight is used to indicate a degree of relevance between the second video and the first video; A feature representation of the first video is constructed based on a weighted average of the feature representations of the n second videos, where the weighted average of the feature representations of the n second videos is determined based on the feature representation of the second video and a feature weight of the second video.

4. The method according to claim 3, characterized in that The obtaining the feature weight of the second video includes: Obtaining similarity information of the second video, wherein the similarity information is used to describe the degree of similarity between the second video and the first video; and determining a feature weight of the second video based on the similarity information of the second video; and / or, Obtain playback information of the second video, wherein the playback information is used to describe the playback status of the second video; and determine a feature weight of the second video based on the playback information of the second video.

5. The method according to any one of claims 1 to 4, characterized in that: The obtaining of the first video and n second videos further includes: Obtaining the first video; Obtain n trailer videos according to the first video.

6. The method according to claim 5, characterized in that The obtaining of n trailer videos according to the first video includes: Acquire a plurality of trailer videos according to the first video; Filter n trailer videos from the multiple trailer videos.

7. The method according to claim 6, characterized in that The step of selecting n trailer videos from the plurality of trailer videos includes: From the plurality of trailer videos, excluding the trailer videos whose feature weights are lower than a first target threshold, to obtain n trailer videos; The feature weight of the trailer video is used to indicate the degree of relevance between the trailer video and the first video.

8. The method according to any one of claims 1 to 4, characterized in that: The recommending n recommended videos includes: Recommending multiple recommended videos, and selecting n recommended videos from the multiple recommended videos.

9. The method according to claim 8, characterized in that The step of selecting n recommended videos from the plurality of recommended videos includes: excluding, from the plurality of recommended videos, the recommended videos whose feature weights are lower than a second target threshold, to obtain n recommended videos; The feature weight of the recommended video is used to indicate the relevance between the recommended video and the first video.

10. The method according to any one of claims 1 to 4, characterized in that: The feature representation of the second video includes at least one of the following feature representations: a feature representation of an image dimension, a feature representation of an audio dimension, and a feature representation of a text dimension.

11. A device for constructing a video feature representation, characterized in that: The device comprises: An acquisition module is configured to acquire a first video and n second videos, wherein the video length of the second videos is shorter than that of the first video, where n is a plurality of videos, and the second videos include recommended videos, or the second videos include a trailer video of the first video and the recommended videos, wherein the recommended videos are videos recommended based on the video attributes of the first video; wherein the first video is acquired, and n recommended videos are recommended based on the first video, and / or the first video and the trailer video of the first video are acquired, and n recommended videos are recommended based on the trailer video; The acquisition module is further configured to acquire feature representations of the n second videos, where the feature representations of the second videos are used to represent hidden layer features of the second videos; A construction module is used to construct a feature representation of the first video based on the feature representation of the second video, where the feature representation of the first video is used to represent the hidden layer features of the first video.

12. A computer device, characterized in that: The computer device includes: a processor and a memory, wherein the memory stores at least one program; the processor is used to execute the at least one program in the memory to implement the method for constructing the video feature representation according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that The computer-readable storage medium stores executable instructions, which are loaded and executed by a processor to implement the method for constructing a video feature representation according to any one of claims 1 to 10.

14. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. A processor reads and executes the computer instructions from the computer-readable storage medium to implement the method for constructing a video feature representation according to any one of claims 1 to 10.

Images