CN109104620B - Short video recommendation method and device and readable medium - Google Patents

Short video recommendation method and device and readable medium Download PDF

Info

Publication number
CN109104620B
CN109104620B CN201810837633.6A CN201810837633A CN109104620B CN 109104620 B CN109104620 B CN 109104620B CN 201810837633 A CN201810837633 A CN 201810837633A CN 109104620 B CN109104620 B CN 109104620B
Authority
CN
China
Prior art keywords
short video
short
model
training
watched
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810837633.6A
Other languages
Chinese (zh)
Other versions
CN109104620A (en
Inventor
刘龙坡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN201810837633.6A priority Critical patent/CN109104620B/en
Publication of CN109104620A publication Critical patent/CN109104620A/en
Application granted granted Critical
Publication of CN109104620B publication Critical patent/CN109104620B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H04N21/25866—Management of end-user data
    • H04N21/25891—Management of end-user data being end-user preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/4508—Management of client data or end-user data
    • H04N21/4532—Management of client data or end-user data involving end-user characteristics, e.g. viewer profile, preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computing Systems (AREA)
  • Computer Graphics (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a short video recommendation method, a device and a readable medium, belonging to the technical field of video recommendation, wherein in the method and the device provided by the invention, after a short video pull request is received, a short video sequence formed by a short video list watched by a user in history and a short video list not watched by the user is obtained; determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features; determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained by training; and recommending interested short videos to the user according to the probability of each short video. By implementing the method, the short videos which are interesting to the user are determined from the mass short videos and are recommended to the user, so that the watching requirements of the user on the short videos are met, and the utilization rate of the short video application program of the user is improved.

Description

Short video recommendation method and device and readable medium
Technical Field
The invention relates to the technical field of video recommendation, in particular to a short video recommendation method, a short video recommendation device and a readable medium.
Background
Currently, as a new video viewing platform, short video applications such as tremble, volcano small video, fast-handed videos, and micro-videos are used, there are many videos and authors, and an effective content organization recommendation method is not available to select interesting content from a large amount of short videos for a user.
Therefore, how to recommend the short videos in which the user is interested from the massive short videos becomes one of the technical problems to be solved urgently in the prior art.
Disclosure of Invention
The embodiment of the invention provides a short video recommendation method, a short video recommendation device and a readable medium, which are used for recommending short videos which are interesting to a user from massive short videos.
In a first aspect, an embodiment of the present invention provides a short video recommendation method, including:
after receiving a short video pulling request, acquiring a short video sequence formed by a short video list watched by a user in history and an unviewed short video list, wherein the short video sequence comprises identification information of each short video;
determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features;
determining the probability of each short video in the unviewed short video list according to the sequence vector and a short video recommendation model obtained by training; and are
And recommending interested short videos to the user according to the probability of each short video.
By utilizing the list of the short videos watched by the user in history, the short videos not watched and the short video recommendation model provided by the invention, the short videos which are interesting to the user can be determined from the mass short videos and recommended to the user.
Preferably, the identification information of the short video is the number of the short video; and determining a sequence vector for characterizing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for characterizing all the short video features, specifically comprising:
for each short video in the short video sequence, extracting a row vector with the same number as that of the short video from the short video feature matrix, and determining the row vector as a video vector of the short video;
and constructing a sequence vector for representing the short video characteristics in the short video sequence by using the video vector obtained based on each short video in the short video sequence.
Preferably, the short video feature matrix used for characterizing all the short video features is obtained by training the embedded embedding learning algorithm by using a short video list viewed by a user in history.
Preferably, the short video recommendation model is trained as follows:
acquiring training samples, wherein the training samples are composed of historically watched short videos, future watched short videos and label results of the future watched short videos, and the future watched short videos are generated in a future time period with an acquisition time point as a reference;
carrying out embedding processing on the training sample to obtain a training feature matrix for representing short video features in the training sample;
and training the short video recommendation model by using the training feature matrix.
Further, the short video recommendation model is composed of two Long Short Term Memory (LSTM) models and a Deep Neural Network (DNN) model; and training the short video recommendation model by using the training feature matrix, specifically comprising:
inputting the training feature matrix into a first-layer LSTM model; and are
Inputting the output result of the first layer of LSTM model into a second layer of LSTM model, and extracting the output result corresponding to the short video for future viewing from the output result of the second layer of LSTM model; and are
Inputting an output result corresponding to the short video for future viewing into a DNN model;
and adjusting the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model according to the output result of the DNN model and the label result of the short video to be watched in the future until the output result of the DNN model and the label result of the short video to be watched in the future reach a model training end condition.
In a second aspect, an embodiment of the present invention provides a short video recommendation apparatus, including:
the short video pull-out device comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is used for acquiring a short video sequence formed by a short video list watched by a user history and an unviewed short video list after receiving a short video pull-out request, and the short video sequence comprises identification information of each short video;
a first determining unit, configured to determine, according to the short video sequence and a short video feature matrix obtained through training and used for representing all short video features, a sequence vector used for representing the short video features in the short video sequence;
the second determining unit is used for determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained through training; and are
And the recommending unit is used for recommending the interested short videos to the user according to the probability of each short video.
Preferably, the identification information of the short video is the number of the short video; and
the first determining unit is specifically configured to, for each short video in the short video sequence, extract a row vector with a same number as that of the short video from the short video feature matrix, and determine the row vector as a video vector of the short video; and constructing a sequence vector for representing the short video characteristics in the short video sequence by using the video vector obtained based on each short video in the short video sequence.
Preferably, the short video feature matrix used for characterizing all the short video features is obtained by training the embedded embedding learning algorithm by using a short video list viewed by a user in history.
Preferably, the apparatus further comprises:
the model training unit is used for training the short video recommendation model according to the following method: acquiring training samples, wherein the training samples are composed of historically watched short videos, future watched short videos and label results of the future watched short videos, and the future watched short videos are generated in a future time period with an acquisition time point as a reference; carrying out embedding processing on the training sample to obtain a training feature matrix for representing short video features in the training sample; and training the short video recommendation model by using the training feature matrix.
Further, the short video recommendation model is composed of two Long Short Term Memory (LSTM) models and a Deep Neural Network (DNN) model; and
the model training unit is specifically used for inputting the training feature matrix into the first-layer LSTM model; inputting the output result of the first layer of LSTM model into a second layer of LSTM model, and extracting the output result corresponding to the short video to be watched in the future from the output result of the second layer of LSTM model; inputting an output result corresponding to the short video to be watched in the future into a DNN model; and adjusting the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model according to the output result of the DNN model and the label result of the short video to be watched in the future until the output result of the DNN model and the label result of the short video to be watched in the future reach a model training end condition.
In a third aspect, an embodiment of the present invention provides a computer-readable medium storing computer-executable instructions for performing the short video recommendation method provided in the present application.
In a fourth aspect, an embodiment of the present invention provides a computing apparatus, including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the short video recommendation methods provided herein.
The invention has the beneficial effects that:
according to the short video recommendation method, device and readable medium provided by the embodiment of the invention, after a short video pull request is received, a short video sequence formed by a short video list watched by a user in history and an unviewed short video list is obtained, wherein the short video sequence comprises identification information of each short video; determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features; determining the probability of each short video in the unviewed short video list according to the sequence vector and a short video recommendation model obtained by training; and recommending interested short videos to the user according to the probability of each short video. By implementing the method, short videos interesting to the user can be determined from a large number of short videos and recommended to the user by utilizing the short video list watched by the user in history, the short videos not watched by the user and the short video recommendation model provided by the invention, so that the watching requirements of the user on the short videos are met, and the utilization rate of the short video application program by the user is improved.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:
fig. 1 is a schematic view of an application scenario of a short video recommendation method according to an embodiment of the present invention;
fig. 2 is a flowchart illustrating a short video recommendation method according to an embodiment of the present invention;
fig. 3 is a second flowchart illustrating a short video recommendation method according to an embodiment of the present invention;
FIG. 4 is a schematic flowchart of determining a sequence vector for characterizing short video features in a short video sequence according to an embodiment of the present invention;
fig. 5 is a schematic flowchart of a process of training a short video recommendation model according to an embodiment of the present invention;
fig. 6 is a second schematic flowchart of a process of training a short video recommendation model according to an embodiment of the present invention;
fig. 7 is a schematic flowchart of a process of training a short video recommendation model by using a training feature matrix according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a short video recommendation apparatus according to an embodiment of the present invention;
fig. 9 is a schematic hardware structure diagram of a computing apparatus for implementing the short video recommendation method according to an embodiment of the present invention.
Detailed Description
The short video recommendation method, device and readable medium provided by the embodiment of the invention are used for recommending the short video which is interested by the user to the user from the massive short videos.
The preferred embodiments of the present invention will be described below with reference to the accompanying drawings of the specification, it being understood that the preferred embodiments described herein are merely for illustrating and explaining the present invention, and are not intended to limit the present invention, and that the embodiments and features of the embodiments in the present invention may be combined with each other without conflict.
To facilitate understanding of the invention, the present invention relates to technical terms in which:
1. embedding is used to convert the input data into a vector with a fixed size. The embedding learning algorithm is trained by utilizing a short video list viewed by a user history, a short video feature matrix for representing all short video features can be obtained, and then the input short video sequence can be converted into a low-dimensional sequence vector for representing the short video features in the short video sequence based on the short video feature matrix for representing all the short video features obtained by training.
2. A Long Short Term Memory model (LSTM), an improved recurrent neural network, can solve the problem that RNN cannot handle Long-distance dependence. The invention can capture the time sequence relation of the short video in the short video sequence by utilizing the two layers of LSTMs, and can better determine the short video which is more interesting to the user in the current time on the basis.
3. Deep Neural Network (DNN) is a Neural Network with at least one hidden layer. Similar to the shallow neural network, the deep neural network can also provide modeling for a complex nonlinear system, but the extra levels provide higher abstraction levels for the model, thereby improving the capability of the model.
4. The short video list watched by the user history is described by taking the total number of the current short videos as 10000 as an example, after the 10000 short videos are numbered, if the user watches the 2 nd, 6 th, 8 th short videos and the like, the short video list watched by the user history is: [2,6,8, … … ].
5. The electronic device may be mobile or fixed, and for example, may be a mobile phone, a tablet computer, various wearable devices, an in-vehicle device, a Personal Digital Assistant (PDA), a point of sale (POS), a device capable of performing short video recommendation, or other electronic devices capable of implementing the above functions.
6. An application program, which is a computer program capable of performing one or more specific tasks, has a visual display interface and can interact with a user, and for example, an electronic map, a WeChat, and the like can be referred to as an application program.
7. In the description of the embodiments of the invention, the terms "first," "second," and the like are used for descriptive purposes only and not for purposes of indicating or implying relative importance, nor for purposes of indicating or implying order.
In order to recommend short videos in which a user is interested from a large number of short videos to the user, an embodiment of the present invention provides a solution, referring to an application scene schematic diagram shown in fig. 1, an application program with a short video playing function is installed on user equipment 11, then the user sends a short video pull request to a server 12 through the application program installed in the user equipment 11, after receiving the short video pull request, the server 12 obtains a short video sequence formed by a short video list watched by the user in history and a short video list not watched by the user, and the short video sequence includes identification information of each short video; determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features; determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained by training; and according to the probability of each short video, determining the short video which is interested by the user, and recommending the short video which is interested by the user to the user through an application program in the user equipment 11, so that the user can play the short video. Therefore, by adopting the method provided by the invention, the short videos which are relatively interested by the user can be determined from the massive short videos by utilizing the short video recommendation model obtained by training according to the short video list watched by the user history and the short video list not watched by the user, so that the requirement of the user on the short videos is met, and the satisfaction degree and the utilization rate of the user on the application program are improved.
The user equipment 11 and the server 12 are communicatively connected through a network, which may be a local area network, a wide area network, or the like. The user equipment 11 may be a portable device (e.g., a mobile phone, a tablet, a notebook Computer, etc.) or a Personal Computer (PC), the server 12 may be any device capable of providing internet services, and the application program in the user equipment 11 may be an application program with a short video playing function, such as a tremble, a fast-hand, a micro-vision, etc.
In the invention, after receiving the short video pulling requests of a plurality of users at the same time, the server 12 can determine interested short videos for the plurality of users and recommend the short videos by executing the method provided by the invention.
It should be noted that, if the user equipment 11 has strong storage capability and processing capability, the short video recommendation method provided by the present invention may also be executed by the user equipment 11 to recommend the short video of interest to the user 10.
The short video recommendation method provided according to the exemplary embodiment of the present invention is described below with reference to the application scenario of fig. 1. It should be noted that the above application scenarios are merely illustrated for the convenience of understanding the spirit and principles of the present invention, and the embodiments of the present invention are not limited in this respect. Rather, embodiments of the present invention may be applied to any scenario where applicable.
As shown in fig. 2, a flowchart of a short video recommendation method provided in an embodiment of the present invention may include the following steps:
and S21, after receiving the short video pulling request, acquiring a short video sequence formed by the short video list watched by the user history and the short video list not watched by the user.
The short video sequence in the invention comprises the identification information of each short video.
In this step, when the user opens the application program with the short video playing function in the user equipment 11, a short video pull request may be triggered, or when the user is watching a short video in the application program, the short video pull request may also be triggered by sliding down the main interface, so that the application program may send the short video pull request to the server 12 after receiving the short video pull request. In order to ensure that the server 12 determines the short videos in which the user is interested, the short video pulling request carries identification information of the user in an application program, such as a user name of the user, so that the server is ensured to determine the short videos in which the user is interested from a large number of short videos for the user by using the short video recommending method provided by the invention and display the short videos to the user through the application program.
Specifically, after receiving the short video pull request, the server 12 may obtain, according to the identification information of the user carried in the short video pull request, a short video list formed by short videos watched by the user in a preset time period before the current time, that is, the short video list historically watched by the user in the present invention, and then the server 12 summarizes all the short videos, and by comparing all the short videos with the short videos in the short video list historically watched by the user, the short video list not watched by the user may be determined. In order to perform the subsequent process, the two short video lists need to be spliced together to obtain a short video sequence, i.e., the input sequence in fig. 3.
It should be noted that all short videos in the present invention can be understood as all short videos generated from the beginning of the short video application to the current time. However, since the short videos are also time-efficient, and users generally prefer the newer and hotter short videos, all the short videos in the present invention can be understood as all the short videos generated within a preset time period before the current time, rather than all the short videos generated from the beginning of the short video application to the current time.
Preferably, for convenience of calculation, only the identification information of each short video, such as the video ID of the short video, may be stored in the short video sequence in the present invention.
And S22, determining a sequence vector for characterizing the short video features in the short video sequence according to the short video sequence and the short video feature matrix obtained by training and used for characterizing all the short video features.
In this step, sequence vectors capable of representing short video features watched and not watched by the user in the short video sequence are determined, so that after the sequence vectors are input into the short video recommendation model, short videos containing the short video features in the non-watched short videos can be predicted based on the short video features of the short videos watched by the user, and the short videos in which the short video features are not watched by the user, namely the short videos in which the user is interested are determined.
In practical applications, since the video ID is generally longer, the identification information of the short video in the present invention may be the number of the short video in order to save processing resources. Specifically, all the short videos can be numbered, it is ensured that each short video corresponds to a unique number, a corresponding relationship between the video ID of the short video and the number of the short video is established, and only the number of the short video is stored in the short video sequence, so that on one hand, calculation is convenient, and on the other hand, the short video corresponding to the number can be quickly and accurately found according to the corresponding relationship.
On this basis, step S22 can be executed according to the method shown in fig. 4, namely, determining a sequence vector for characterizing the short video features in the short video sequence of the present invention, including the following steps:
and S31, extracting a row vector with the same number as that of the short video from the short video feature matrix for each short video in the short video sequence, and determining the row vector as a video vector of the short video.
In this step, the short video feature matrix is taken as Mm*nShort video sequences are [ X1, X2, X3, X4, X5, X6, … …%]For illustration, where m is largeEqual to the number of all short videos. Then for short video Xi, then from short video feature matrix Mm*nThe line vector corresponding to the Xi-th line is extracted, and then the line vector is determined as the video vector Li of the short video Xi. Based on the above description, a video vector for each short video in the short video sequence, i.e., the video vector in fig. 3, can be determined. The values of m and n in the present invention may be determined according to actual conditions, and the present invention is not limited to the values.
And S32, forming a sequence vector for representing the short video characteristics in the short video sequence by the video vector obtained based on each short video in the short video sequence.
In this step, as described with reference to fig. 3, the video vectors of each short video in the short video sequence obtained in step S31 may be spliced to obtain a sequence vector capable of representing the short video feature in the short video sequence. For example, the video vectors of the short videos in the short video sequence obtained in step S31 are: l1, L2, L3, L4, L5, L6, … …, the sequence vector based on this may be [ L1, L2, L3, L4, L5, L6, … … [ ]]TThe sequence vector is the sequence vector in fig. 3.
Preferably, the short video feature matrix used for characterizing all the short video features in the present invention is obtained by training the embedded embedding learning algorithm by using the short video list viewed by the user history.
Specifically, after the short videos of the application program are numbered, the number of the short videos watched by the user in the recent period of time can be obtained to form a short video list watched by the user in history, and assuming that the total number of the current short videos is 10000, the short videos watched by the user in history are: 2,4,6,8,10,12,14,20,40,45,56,66, … …. Because the number of short videos watched by each user in a preset time period before the current time is inconsistent and may be large, the length of the user history watching list is limited, for example, the short videos watched by each user history are sorted according to time, the number of k short videos closest to the current time is taken to form the short video list watched by the user history, and taking k as an example for explanation, the short video list formed by the short videos watched by the user history is [2,4,6,8,10,12,14,20,40,45,56,66 ]. It should be noted that, if the amount of short video frequency watched by the user within the preset time period is less than k, a 0 is complemented at the end in the short video list watched by the user in history. The length of the short video list watched by each user is consistent.
Based on the description, the short video lists watched by each user in history can be obtained, and then the short video lists of each user are used for sequencing the embedding learning algorithm, so that the embedding learning algorithm can learn the features of each short video, and further the short video feature matrix used for representing the short video features of all the short videos is obtained. When all short videos are 10000, the obtained short video feature matrix for representing all short video features is 10000 × 128 matrix, namely Mm*n＝M10000*128。
After the short video feature matrix is obtained, when the short video sequence composed of the short video viewed and the short video not viewed based on the user history acquired in step S11 is [3,5,7,12,20,32,45,46,56,58,78 ]]If the number 78 is the unviewed short video, the process of obtaining the sequence vector in step S32 is as follows: the total number of the short videos is 10000, and the short video feature matrix is M10000*128For example, since the total number of short videos is 10000, the feature space of the short video list viewed by the user in history is 10000 dimensions, and the matrix expansion is very sparse, the short video list can be mapped into a low-dimensional matrix, for example, a matrix with dimension 128, by using the embedding technique, that is, the short video list viewed by the user in history is [3,5,7,12,20,32,45,46,56,58]Then, the embedding technique can be used to obtain a 10 × 128 low-dimensional matrix. Specifically, the user original matrix is expanded into a sparse matrix of 10 × 10000, and the sparse matrix is trained by an embedding technology and then embedded into a low-dimensional dense matrix of 10 × 128. The mapping process is as follows: learning a 10000 × 128 short video feature matrix M through the embedding technology, and obtaining a list of historically viewed short videos Mu＝[3,5,7,12,20,32,45,46,56,58,78]TI.e. MuIs a matrix of 10 x1, then for the short video matrix numbered 3, then from the short video feature matrix M10000*128Extracting the 3 rd row of data to obtain video vectors for representing the characteristics of the short videos with the number 3, so that the video vectors of the short videos with the numbers in the short video list historically watched by the user can be obtained, namely a matrix of 10 × 128 is formed; then, for the short video 78 that is not viewed, the 78 th row of data is extracted from the short video feature matrix M to obtain a video vector representing the features of the short video of number 78, and the video vector and the matrix of 10 × 128 are formed into a sequence vector of 11 × 128.
And S23, determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained by training.
In this step, the probability of each short video in the short video list not watched can be determined by inputting the short videos including the short videos historically watched by the user and the short videos not watched by the user into the short video recommendation model trained in advance, so that the server 12 can send the short videos with higher probability to the user equipment 11 of the user, preferentially recommend the short videos to the user based on the application program in the user equipment 11, and determine the short videos in which the user is interested from a large number of short videos. Specifically, after obtaining the sequence vector based on step S22, the sequence vector may be input to the short video recommendation model in fig. 3, so as to obtain the probability of the short video not being viewed.
Preferably, the short video recommendation model in the present invention is obtained by training according to the procedure shown in fig. 5, and may include the following steps:
and S41, obtaining a training sample.
The training samples in the invention are formed by label results of historically watched short videos, future watched short videos and future watched short videos, wherein the future watched short videos in the invention are short videos generated in a future time period by taking a collection time point as a reference.
In this step, the training samples in the present invention are collected, for each user, short videos that have been watched by the user for a period of time in the future and have never been watched in the past as positive samples, and short videos that have not been watched by the user for a period of time in the future and have never been watched in the past as negative samples. Based on this principle, tags are added to future view short videos. The training sample is obtained by splicing a historical short video list and a future short video list and a label result thereof. For example, the short video sequence that the user has historically viewed is: [3,5,7,12,20,32,45,46,56,58], the short video is viewed in the future at 78, if the user has viewed the short video numbered 78, the short video is a positive sample, i.e., the labeling result of the short video numbered 78 is 1, and if the user has not viewed the short video, the short video is indicated as a negative sample, i.e., the labeling result of the short video numbered 78 is 0.
Specifically, based on a certain collection time point, a list of short videos historically watched by the user and a list of short videos watched in the future are obtained, that is, on the basis of 10:00, the short videos watched by the user between 9:30 and 10:00 can be determined as the short videos historically watched by the user, for example, [3,5,7,12,20,32,45,46,56,58]Determining short videos between 10:00 and 10:30 as unviewed short videos, randomly extracting one short video from a short video list between 10:00 and 10:30, for example, if the extracted short video is 78, determining the label result of the short video 78, and comparing [3,5,7,12,20,32,45,46,56,58, 32]And 78 and the label result of the number 78 are spliced to obtain a training sample, namely the training sample is as follows: [3,5,7,12,20,32,45,46,56,58,78Labeling results]。
And S42, performing embedding processing on the training samples to obtain a training feature matrix for representing the short video features in the training samples.
In this step, the training samples obtained in step S41 are subjected to embedding dimensionality reduction processing to obtain a training feature matrix of 11 × 128, that is, a short video feature matrix M10000*128The data of the rows 3,5,7,12,20,32,45,46,56,58 and 78 are extracted and converted into a training feature matrix of 11 × 128.
And S43, training the short video recommendation model by using the training feature matrix.
In this step, the training feature matrix obtained based on step S42 may be input into a short video model for training, so as to obtain a short video recommendation model for implementing a short video recommendation function, so as to implement recommendation of a short video that is of interest to a user by using the method provided by the present invention.
Preferably, the short video recommendation model in the invention is composed of two long-short term memory (LSTM) models and a Deep Neural Network (DNN) model; referring to the schematic structural diagram of the short video recommendation model shown in fig. 6, step S43 may be executed according to the method shown in fig. 7, and may include the following steps:
and S51, inputting the training feature matrix into the first-layer LSTM model.
In this step, the parameter configuration of each parameter of the first layer LSTM model may refer to table 1:
TABLE 1
Parameter configuration Parameter value
Number of neurons 64
Length of input sequence 11
Activating a function relu
dropout coefficient 0.5
By setting the first layer LSTM model according to the above parameters, overfitting can be prevented. The length of the input sequence in the present invention is set to 11, or may be set to another value, and the length of the input sequence is 11 for example, then the training sample in step S41 may be first input into the embedding learning algorithm in fig. 6 to obtain a training feature matrix of 11 × 128, and then the training feature matrix may be input into the first layer LSTM model in fig. 6, so that the training feature matrix of 11 × 128 may be converted into a matrix of 11 × 64, and at the same time, the time sequence relationship between the 11 short videos may be captured.
And S52, inputting the output result of the first-layer LSTM model into the second-layer LSTM model, and extracting the output result corresponding to the short video to be watched in the future from the output result of the second-layer LSTM model.
In this step, the parameter configuration of each parameter of the second layer LSTM model may refer to table 2:
TABLE 2
Parameter configuration Parameter value
Number of neurons 32
Length of input sequence 11
Activating a function relu
dropout coefficient 0.5
By setting the second layer LSTM model according to the above parameters, overfitting can be prevented. The dimension can be reduced from 64 to 32 by inputting the output of the first layer LSTM model, a matrix of 11 x 64, into the second layer LSTM model in fig. 6. However, at this time, the matrix of 11 × 32 is not output, but only the output result of the unviewed short video represented by the 11 th row is output, and after the data represented by the 11 th row is input to 32 neurons, each neuron can obtain one result, so that a matrix of 32 × 1 can be obtained to represent the output result corresponding to the unviewed short video.
And S53, inputting the output result corresponding to the short video to be watched in the future into the DNN model.
In this step, the parameter configuration of each parameter of the DNN model may refer to table 3:
TABLE 3
Parameter configuration Parameter value
Number of neurons 1
Length of input sequence 32
Activating a function sigmod
Output dimension 1
And ensuring that the output result is between [0 and 1] by setting the DNN model according to the parameters. After setting the DNN model according to the above parameters, by inputting the matrix 32 × 1 obtained in step S53 into the DNN model shown in fig. 6, the probability of an unviewed short video, that is, the prediction score of the model on whether the user will view the 11 th video in the future, can be obtained.
S54, according to the output result of the DNN model and the label result of the short video to be watched in the future, the weights of the neurons in the embedding learning algorithm, the first-layer LSTM model, the second-layer LSTM model and/or the DNN model are adjusted until the output result of the DNN model and the label result of the short video to be watched in the future reach the model training end condition.
In this step, because the output result of the DNN model represents the probability of whether the user watches the short video in the future, the output result is compared with the label result of watching the short video in the future, and if the output result and the label result meet the condition of the model training result, it indicates that the current model does not need to be adjusted, and the short video recommendation can be directly performed; otherwise, the current model recommendation result is incorrect, and the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model need to be adjusted. For example, if the output result and the tag result are within an allowable difference range, the model is available; if the output result and the label result are not in the allowable difference range, the model prediction result is inaccurate and needs to be adjusted. For example, the output result is 0.988, the tag result is 1, it can be seen that the output result is very close to the tag result, and within the allowable difference range, it indicates that the model is usable, and the short video recommendation can be directly performed without adjustment.
For example, if the output result is 0.8 and the tag result is 1, it can be seen that the difference between the output result and the tag result is large, indicating that the model recommendation result is incorrect, the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model are adjusted to ensure that the output result of the DNN model and the tag result of the short video viewed in the future reach the model training end condition.
In practical applications, a large number of training samples may be required to train the short video model, and the present invention is not described herein.
It should be noted that, parameters in the two-layer LSTM and DNN models in the present invention may be adjusted according to actual situations, for example, the length of the input sequence is adjusted, if the length of the input sequence is 11, the probability of only one unviewed short video can be predicted each time, and the method provided by the present invention needs to be executed multiple times to determine the probability of each short video in the unviewed short video list; if the length of the input sequence is enough, the probability of all short videos in the short video list which is not watched can be predicted only by executing once, and the parameters of the short video recommendation model can be adjusted according to the actual situation in the specific implementation.
The trained embedding, two-layer LSTM and DNN models can be obtained by performing steps S51 to S54, and when step S23 is performed based on these models, after the sequence vector is obtained based on step S22, the probability of each unviewed short video in the unviewed short video list can be obtained by inputting the sequence vector into the first-layer LSTM, second-layer LSTM and DNN models.
And S24, recommending interested short videos to the user according to the probability of each short video.
In this step, the server 12 may sort the short videos in the unviewed short video list according to the order of the probabilities from high to low based on the probabilities of the short videos in the unviewed short video list determined in step S23, and then recommend the short videos in which the user is interested to the user through the application program in the user equipment 11 according to the sort.
Preferably, after the sequence vector is determined, the present invention may obtain the user vector of the user by averaging according to the rows, then perform similarity calculation using the user vector and each video vector obtained in step S31, and then perform short video recommendation to the user according to the calculation result. Of course, the user vector and each video vector can also be subjected to clustering processing, and then the short videos which are interested by the user can be recommended to the user according to the processing result.
The short video recommendation method provided by the invention comprises the steps of obtaining a short video sequence formed by a short video list watched by a user in history and an unviewed short video list after receiving a short video pull request; determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features; determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained by training; and recommending interested short videos to the user according to the probability of each short video. By implementing the method, short videos interesting to the user can be determined from a large number of short videos and recommended to the user by utilizing the short video list watched by the user in history, the short videos not watched by the user and the short video recommendation model provided by the invention, so that the watching requirements of the user on the short videos are met, and the utilization rate of the short video application program by the user is improved.
Based on the same inventive concept, the embodiment of the invention also provides a short video recommendation device, and as the principle of the device for solving the problems is similar to the short video recommendation method, the implementation of the device can refer to the implementation of the method, and repeated parts are not described again.
As shown in fig. 8, a schematic structural diagram of a short video recommendation apparatus provided in an embodiment of the present invention includes:
the acquiring unit 61 is configured to acquire, after receiving a short video pull request, a short video sequence formed by a short video list historically watched by a user and an unviewed short video list, where the short video sequence includes identification information of each short video;
a first determining unit 62, configured to determine, according to the short video sequence and a short video feature matrix obtained through training and used for representing all short video features, a sequence vector used for representing the short video features in the short video sequence;
a second determining unit 63, configured to determine, according to the sequence vector and a trained short video recommendation model, a probability of each short video in the short video list that is not watched; and are
And the recommending unit 64 is used for recommending the interested short videos to the user according to the probability of each short video.
Preferably, the identification information of the short video is the number of the short video; and
the first determining unit 62 is specifically configured to, for each short video in the short video sequence, extract a row vector with a same number as that of the short video from the short video feature matrix, and determine the row vector as a video vector of the short video; and constructing a sequence vector for representing the short video characteristics in the short video sequence by using the video vector obtained based on each short video in the short video sequence.
Preferably, the short video feature matrix used for characterizing all the short video features is obtained by training the embedded embedding learning algorithm by using a short video list viewed by a user in history.
Preferably, the apparatus further comprises:
the model training unit is used for training the short video recommendation model according to the following method: acquiring training samples, wherein the training samples are composed of historically watched short videos, future watched short videos and label results of the future watched short videos, and the future watched short videos are generated in a future time period with an acquisition time point as a reference; carrying out embedding processing on the training sample to obtain a training feature matrix for representing short video features in the training sample; and training the short video recommendation model by using the training feature matrix.
Further, the short video recommendation model is composed of two Long Short Term Memory (LSTM) models and a Deep Neural Network (DNN) model; and
the model training unit is specifically used for inputting the training feature matrix into the first-layer LSTM model; inputting the output result of the first layer of LSTM model into a second layer of LSTM model, and extracting the output result corresponding to the short video to be watched in the future from the output result of the second layer of LSTM model; inputting an output result corresponding to the short video to be watched in the future into a DNN model; and adjusting the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model according to the output result of the DNN model and the label result of the short video to be watched in the future until the output result of the DNN model and the label result of the short video to be watched in the future reach a model training end condition.
For convenience of description, the above parts are separately described as modules (or units) according to functional division. Of course, the functionality of the various modules (or units) may be implemented in the same or in multiple pieces of software or hardware in practicing the invention.
Having described the short video recommendation method, apparatus, and readable medium of exemplary embodiments of the present invention, a computing apparatus according to another exemplary embodiment of the present invention is next described.
As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or program product. Thus, various aspects of the invention may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.
In some possible embodiments, a computing device according to the present invention may comprise at least one processing unit, and at least one memory unit. Wherein the storage unit stores program code, which, when executed by the processing unit, causes the processing unit to perform the steps of the short video recommendation method according to various exemplary embodiments of the present invention described above in this specification. For example, the processing unit may perform the short video recommendation flow in steps S11 to S14 as shown in fig. 2.
The computing device 70 according to this embodiment of the invention is described below with reference to fig. 9. The computing device 70 shown in fig. 9 is only an example and should not impose any limitations on the functionality or scope of use of embodiments of the present invention.
As shown in fig. 9, the computing apparatus 70 is embodied in the form of a general purpose computing device. Components of computing device 70 may include, but are not limited to: the at least one processing unit 71, the at least one memory unit 72, and a bus 73 connecting the various system components (including the memory unit 72 and the processing unit 71).
Bus 73 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a processor, or a local bus using any of a variety of bus architectures.
The storage unit 72 may include readable media in the form of volatile memory, such as Random Access Memory (RAM)721 and/or cache memory 722, and may further include Read Only Memory (ROM) 723.
The memory unit 72 may also include a program/utility 725 having a set (at least one) of program modules 724, such program modules 724 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
Computing device 70 may also communicate with one or more external devices 74 (e.g., keyboard, pointing device, etc.), with one or more devices that enable a user to interact with computing device 70, and/or with any devices (e.g., router, modem, etc.) that enable computing device 70 to communicate with one or more other computing devices. Such communication may be through an input/output (I/O) interface 75. Also, computing device 70 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) through network adapter 76. As shown, network adapter 76 communicates with other modules for computing device 70 over bus 73. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with computing device 70, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.
In some possible embodiments, various aspects of the short video recommendation method provided by the present invention may also be implemented in a form of a program product, which includes program code for causing a computer device to perform the steps of the short video recommendation method according to various exemplary embodiments of the present invention described above in this specification when the program product runs on the computer device, for example, the computer device may perform the short video recommendation procedure in steps S11 to S14 shown in fig. 2.
The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The program product for the short video recommendation method of the embodiments of the present invention may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a computing device. However, the program product of the present invention is not limited in this regard and, in the present document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device over any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., over the internet using an internet service provider).
It should be noted that although several units or sub-units of the apparatus are mentioned in the above detailed description, such division is merely exemplary and not mandatory. Indeed, the features and functions of two or more of the units described above may be embodied in one unit, according to embodiments of the invention. Conversely, the features and functions of one unit described above may be further divided into embodiments by a plurality of units.
Moreover, while the operations of the method of the invention are depicted in the drawings in a particular order, this does not require or imply that the operations must be performed in this particular order, or that all of the illustrated operations must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (12)

1. A short video recommendation method, comprising:
after receiving a short video pulling request, acquiring a short video sequence formed by a short video list watched by a user in history and an unviewed short video list, wherein the short video sequence comprises identification information of each short video;
determining a sequence vector for representing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for representing all the short video features;
determining the probability of each short video in the unviewed short video list according to the sequence vector and a short video recommendation model obtained by training; the short video recommendation model is composed of two long short-term memory (LSTM) models and a Deep Neural Network (DNN) model; the training samples of the short video recommendation model are formed by historically watched short videos, future watched short videos and label results of the future watched short videos, wherein the future watched short videos are short videos generated in a future time period with an acquisition time point as a reference; the model training end condition of the short video recommendation model is that an output result and a label result of the short video to be watched in the future are within an allowed gap range;
and recommending interested short videos to the user according to the probability of each short video.
2. The method according to claim 1, wherein the identification information of the short video is a number of the short video; and determining a sequence vector for characterizing the short video features in the short video sequence according to the short video sequence and a short video feature matrix obtained by training and used for characterizing all the short video features, specifically comprising:
for each short video in the short video sequence, extracting a row vector with the same number as that of the short video from the short video feature matrix, and determining the row vector as a video vector of the short video;
and constructing a sequence vector for representing the short video characteristics in the short video sequence by using the video vector obtained based on each short video in the short video sequence.
3. The method of claim 2, wherein the short video feature matrix for characterizing all short video features is a list of short videos historically viewed by each user, trained using an embedded embedding learning algorithm.
4. The method of claim 1, wherein the short video recommendation model is trained in accordance with the following method:
obtaining a training sample;
carrying out embedding processing on the training sample to obtain a training feature matrix for representing short video features in the training sample;
and training the short video recommendation model by using the training feature matrix.
5. The method of claim 4, wherein training the short video recommendation model using the training feature matrix specifically comprises:
inputting the training feature matrix into a first-layer LSTM model; and are
Inputting the output result of the first layer of LSTM model into a second layer of LSTM model, and extracting the output result corresponding to the short video for future viewing from the output result of the second layer of LSTM model; and are
Inputting an output result corresponding to the short video for future viewing into a DNN model;
and adjusting the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model according to the output result of the DNN model and the label result of the short video to be watched in the future until the output result of the DNN model and the label result of the short video to be watched in the future reach a model training end condition.
6. A short video recommendation device, comprising:
the short video pull-out device comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is used for acquiring a short video sequence formed by a short video list watched by a user history and an unviewed short video list after receiving a short video pull-out request, and the short video sequence comprises identification information of each short video;
a first determining unit, configured to determine, according to the short video sequence and a short video feature matrix obtained through training and used for representing all short video features, a sequence vector used for representing the short video features in the short video sequence;
the second determining unit is used for determining the probability of each short video in the short video list which is not watched according to the sequence vector and the short video recommendation model obtained through training; the short video recommendation model is composed of two long short-term memory (LSTM) models and a Deep Neural Network (DNN) model; the training samples of the short video recommendation model are formed by historically watched short videos, future watched short videos and label results of the future watched short videos, wherein the future watched short videos are short videos generated in a future time period with an acquisition time point as a reference; the model training end condition of the short video recommendation model is that an output result and a label result of the short video to be watched in the future are within an allowed gap range;
and the recommending unit is used for recommending the interested short videos to the user according to the probability of each short video.
7. The apparatus of claim 6, wherein the identification information of the short video is a number of the short video; and
the first determining unit is specifically configured to, for each short video in the short video sequence, extract a row vector with a same number as that of the short video from the short video feature matrix, and determine the row vector as a video vector of the short video; and constructing a sequence vector for representing the short video characteristics in the short video sequence by using the video vector obtained based on each short video in the short video sequence.
8. The apparatus of claim 7, wherein the matrix of short video features used to characterize all short video features is a list of short videos historically viewed by each user, trained using an embedded embedding learning algorithm.
9. The apparatus of claim 6, further comprising:
the model training unit is used for training the short video recommendation model according to the following method: obtaining a training sample; carrying out embedding processing on the training sample to obtain a training feature matrix for representing short video features in the training sample; and training the short video recommendation model by using the training feature matrix.
10. The apparatus of claim 9, wherein the model training unit is specifically configured to: inputting the training feature matrix into a first-layer LSTM model; inputting the output result of the first layer of LSTM model into a second layer of LSTM model, and extracting the output result corresponding to the short video to be watched in the future from the output result of the second layer of LSTM model; inputting an output result corresponding to the short video to be watched in the future into a DNN model; and adjusting the weights of the neurons in the embedding learning algorithm, the first layer LSTM model, the second layer LSTM model and/or the DNN model according to the output result of the DNN model and the label result of the short video to be watched in the future until the output result of the DNN model and the label result of the short video to be watched in the future reach a model training end condition.
11. A computer-readable medium having stored thereon computer-executable instructions for performing the method of any one of claims 1 to 5.
12. A computing device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 5.
CN201810837633.6A 2018-07-26 2018-07-26 Short video recommendation method and device and readable medium Active CN109104620B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810837633.6A CN109104620B (en) 2018-07-26 2018-07-26 Short video recommendation method and device and readable medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810837633.6A CN109104620B (en) 2018-07-26 2018-07-26 Short video recommendation method and device and readable medium

Publications (2)

Publication Number Publication Date
CN109104620A CN109104620A (en) 2018-12-28
CN109104620B true CN109104620B (en) 2020-05-19

Family

ID=64847533

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810837633.6A Active CN109104620B (en) 2018-07-26 2018-07-26 Short video recommendation method and device and readable medium

Country Status (1)

Country Link
CN (1) CN109104620B (en)

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111385659B (en) * 2018-12-29 2021-08-17 广州市百果园信息技术有限公司 Video recommendation method, device, equipment and storage medium
CN109558514B (en) * 2019-01-08 2023-04-11 青岛聚看云科技有限公司 Video recommendation method, device thereof, information processing equipment and storage medium
CN109871464B (en) * 2019-01-17 2020-12-25 东南大学 Video recommendation method and device based on UCL semantic indexing
CN109874053B (en) * 2019-02-21 2021-10-22 南京航空航天大学 Short video recommendation method based on video content understanding and user dynamic interest
CN109871487B (en) * 2019-02-22 2021-03-23 北京搜狐新媒体信息技术有限公司 News recall method and system
CN109996122B (en) * 2019-04-12 2021-07-20 北京奇艺世纪科技有限公司 Video recommendation method and device, server and storage medium
CN110012356B (en) * 2019-04-16 2020-07-10 腾讯科技（深圳）有限公司 Video recommendation method, device and equipment and computer storage medium
CN111858970B (en) * 2019-04-30 2024-01-02 北京达佳互联信息技术有限公司 Multimedia content recommendation method and device, electronic equipment and readable storage medium
CN111898018B (en) * 2019-05-06 2024-05-14 北京达佳互联信息技术有限公司 Virtual resource sending method and device, electronic equipment and storage medium
CN110334245B (en) * 2019-05-20 2023-04-07 山东大学 Short video analysis method and device of graph neural network based on time sequence attribute
CN112000872A (en) * 2019-05-27 2020-11-27 北京地平线机器人技术研发有限公司 Recommendation method based on user vector, training method of model and training device
CN110213671B (en) * 2019-06-04 2021-12-17 北京奇艺世纪科技有限公司 Hotspot short video determination method and device
CN110287412B (en) * 2019-06-10 2023-10-24 腾讯科技（深圳）有限公司 Content recommendation method, recommendation model generation method, device, and storage medium
CN112307313B (en) * 2019-07-31 2024-04-16 北京达佳互联信息技术有限公司 Information recommendation method, device and storage medium
CN110765352B (en) * 2019-10-11 2022-11-11 上海上湖信息技术有限公司 User interest identification method and device
CN111210335B (en) * 2019-12-16 2023-11-14 北京淇瑀信息科技有限公司 User risk identification method and device and electronic equipment
CN111225255B (en) * 2020-01-20 2022-03-15 北京字节跳动网络技术有限公司 Target video push playing method and device, electronic equipment and storage medium
CN112153423A (en) * 2020-09-15 2020-12-29 济南雪景网络技术有限公司 User-self-driven short video intelligent recommendation method, electronic equipment and storage medium
CN112822526B (en) * 2020-12-30 2023-07-07 咪咕文化科技有限公司 Video recommendation method, server and readable storage medium
CN113038163B (en) * 2021-03-26 2023-06-23 百果园技术(新加坡)有限公司 User experience model training method, short video user experience assessment method and device
CN113553505A (en) * 2021-07-23 2021-10-26 咪咕文化科技有限公司 Video recommendation method and device and computing equipment
CN113766281B (en) * 2021-09-10 2022-11-22 北京快来文化传播集团有限公司 Short video recommendation method, electronic device and computer-readable storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108170529A (en) * 2017-12-26 2018-06-15 北京工业大学 A kind of cloud data center load predicting method based on shot and long term memory network
CN108228835A (en) * 2018-01-04 2018-06-29 百度在线网络技术（北京）有限公司 For handling the method and apparatus of video

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103714130B (en) * 2013-12-12 2017-08-22 深圳先进技术研究院 Video recommendation system and method
CN106331778B (en) * 2015-07-06 2020-08-14 腾讯科技（深圳）有限公司 Video recommendation method and device
US10242266B2 (en) * 2016-03-02 2019-03-26 Mitsubishi Electric Research Laboratories, Inc. Method and system for detecting actions in videos
CN105975641A (en) * 2016-07-15 2016-09-28 合网络技术（北京）有限公司 Video recommendation method ad device
CN108260008A (en) * 2018-02-11 2018-07-06 北京未来媒体科技股份有限公司 A kind of video recommendation method, device and electronic equipment

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108170529A (en) * 2017-12-26 2018-06-15 北京工业大学 A kind of cloud data center load predicting method based on shot and long term memory network
CN108228835A (en) * 2018-01-04 2018-06-29 百度在线网络技术（北京）有限公司 For handling the method and apparatus of video

Also Published As

Publication number Publication date
CN109104620A (en) 2018-12-28

Similar Documents

Publication Publication Date Title
CN109104620B (en) Short video recommendation method and device and readable medium
CN109460514B (en) Method and device for pushing information
CN110458107B (en) Method and device for image recognition
CN109492772B (en) Method and device for generating information
JP2022058915A (en) Method and device for training image recognition model, method and device for recognizing image, electronic device, storage medium, and computer program
CN108960316B (en) Method and apparatus for generating a model
WO2023040506A1 (en) Model-based data processing method and apparatus, electronic device, computer-readable storage medium, and computer program product
CN109582825B (en) Method and apparatus for generating information
CN111078940B (en) Image processing method, device, computer storage medium and electronic equipment
US11677693B2 (en) Systems and methods for generating dynamic conversational responses using ensemble prediction based on a plurality of machine learning models
CN112766284B (en) Image recognition method and device, storage medium and electronic equipment
CN110717597A (en) Method and device for acquiring time sequence characteristics by using machine learning model
CN116050496A (en) Determination method and device, medium and equipment of picture description information generation model
CN113033707B (en) Video classification method and device, readable medium and electronic equipment
CN113140012B (en) Image processing method, device, medium and electronic equipment
CN116578925B (en) Behavior prediction method, device and storage medium based on feature images
GB2600817A (en) Systems and methods for generating dynamic interface options using machine learning models
CN117391466A (en) Novel early warning method and system for contradictory dispute cases
CN116863116A (en) Image recognition method, device, equipment and medium based on artificial intelligence
CN115576789A (en) Method and system for identifying lost user
CN113780318B (en) Method, device, server and medium for generating prompt information
CN114357242A (en) Training evaluation method and device based on recall model, equipment and storage medium
CN114638973A (en) Target image detection method and image detection model training method
CN114580533A (en) Method, apparatus, device, medium, and program product for training feature extraction model
CN113742593A (en) Method and device for pushing information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant