[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN115905617A - Video scoring prediction method based on deep neural network and double regularization - Google Patents

Video scoring prediction method based on deep neural network and double regularization Download PDF

Info

Publication number
CN115905617A
CN115905617A CN202310187456.2A CN202310187456A CN115905617A CN 115905617 A CN115905617 A CN 115905617A CN 202310187456 A CN202310187456 A CN 202310187456A CN 115905617 A CN115905617 A CN 115905617A
Authority
CN
China
Prior art keywords
user
video
matrix
users
score
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.)
Granted
Application number
CN202310187456.2A
Other languages
Chinese (zh)
Other versions
CN115905617B (en
Inventor
赵学健
张晶晶
孙知信
孙哲
曹亚东
宫婧
汪胡青
胡冰
徐玉华
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.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
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 Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202310187456.2A priority Critical patent/CN115905617B/en
Publication of CN115905617A publication Critical patent/CN115905617A/en
Application granted granted Critical
Publication of CN115905617B publication Critical patent/CN115905617B/en
Priority to PCT/CN2023/135369 priority patent/WO2024179089A1/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/73—Querying
    • G06F16/735—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9536—Search customisation based on social or collaborative filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0499—Feedforward networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
    • 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
    • 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
    • H04N21/47—End-user applications
    • H04N21/475—End-user interface for inputting end-user data, e.g. personal identification number [PIN], preference data
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Human Computer Interaction (AREA)
  • Library & Information Science (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a video scoring prediction method based on a deep neural network and double regularization, which comprises the steps of reconstructing a user-video scoring matrix, introducing a video associated regular item fused with user activity and a reliable nearest neighbor regular item, constructing a matrix decomposition recommendation model fused with the video associated regular item fused with the user activity and the reliable nearest neighbor regular item, inputting potential features into the deep neural network to obtain a result of a deep neural network model, and combining the result of the deep neural network model with a matrix decomposition structure to obtain a final prediction score, so that the precision of the prediction score is improved; and mining related information in user video comments by using an LDA (latent Dirichlet Allocation) model to generate a user type latent feature matrix and a video type latent feature matrix, combining the user type latent feature matrix and the video type latent feature matrix to obtain a hidden information matrix, and combining the hidden information matrix and an original user video scoring matrix to generate a new user-video scoring matrix, so that the problems of cold start and data sparsity are solved.

Description

Video scoring prediction method based on deep neural network and double regularization
Technical Field
The invention relates to a video scoring prediction method based on a deep neural network and double regularization, and belongs to the field of scoring prediction.
Background
With the rapid development of the internet technology, video resources in each network platform are more and more, abundant video resources are provided for users, and when more choices are provided for the users, troubles and puzzles are brought to the users, and the huge video resources not only increase the difficulty of searching favorite videos of the users, but also lead the process of searching the videos to be very time-consuming. In order to solve the information overload problem, the personalized recommendation system becomes an effective tool for solving the problem. The score prediction is again an important component of the recommendation algorithm. The existing recommendation algorithms mainly have three main categories: collaborative filtering based recommendation algorithms, content based recommendation algorithms, and hybrid recommendation algorithms. At present, the most used collaborative filtering-based recommendation algorithm is the collaborative filtering-based recommendation algorithm, and the most used collaborative filtering-based recommendation algorithm is the model-based collaborative filtering recommendation algorithm, and several algorithms which are relatively common in the model-based collaborative filtering recommendation algorithm include: matrix decomposition model, singular value decomposition, cluster analysis, etc. However, the existing collaborative filtering recommendation algorithm has problems of data sparseness, cold start and the like, which causes inaccurate score prediction of recommended video resources, thereby affecting personalized recommendation results, improving accuracy of prediction scoring of video resources, and further improving recommendation precision becomes one of the hot spots of current research.
Disclosure of Invention
The invention aims to provide a video scoring prediction method based on a deep neural network and double regularization aiming at the defects of the prior art, a user-video scoring matrix is reconstructed, a video association regular term and a reliable nearest neighbor regular term which are fused with user liveness are introduced during matrix decomposition to restrict the learning of a potential feature matrix, the deep neural network is introduced, the nonlinear feature of the deep neural network is utilized to relieve the limitation of linear dot products in the matrix decomposition process, the result of a deep neural network model is combined with the result of double regularization matrix decomposition, and the precision of video scoring prediction is improved.
The technical scheme adopted by the invention is as follows: a video scoring prediction method based on a deep neural network and double regularization is used for improving the precision of scoring prediction of recommended videos, and specifically comprises the following steps:
step S1: processing the video comments, excavating hidden information, combining a hidden information matrix with an original user-video scoring matrix to generate a new user-video scoring matrix, and entering the step S2;
step S2: the method comprises the following steps of adding a biregular term constraint potential feature matrix into a user-video scoring matrix for learning when decomposing a user-video scoring matrix, wherein each user scores a video to make a certain contribution to video similarity, and the user contributions are not completely the same, so that the users can be divided into active users and inactive users according to the liveness of the users, the active users refer to users who score and record a large number of videos, the inactive users refer to users who score and record only a small number of videos, so that the contributions of the active users and the inactive users are separated when the video similarity is calculated, and the liveness of the users is defined as:
Figure SMS_1
equation 1
In the formula 1, the first and second groups of the compound,
Figure SMS_2
the total score of the user u is represented, so that the video similarity calculation method obtained by combining the activity coefficient of the user and the modified cosine similarity comprises the following steps: />
Figure SMS_3
Equation 2
In the formula 2, the first and second groups,
Figure SMS_4
represents the rating of user u on video i, </or>
Figure SMS_5
Represents the rating of user u on video j, and>
Figure SMS_6
represents a rating score for user u, and->
Figure SMS_7
Representing a set of users who have simultaneously scored videos i and j; and (3) introducing the learning of the video associated regularization term constraint project potential feature matrix integrated with the user activity during matrix decomposition, wherein the video associated regularization constraint function formula integrated with the user activity at the moment is as follows:
Figure SMS_8
equation 3
In equation 3, V represents a video feature matrix, V j Is a potential feature vector of video j，V i Is a potential feature vector of video i and proceeds to step S3;
and step S3: taking the potential characteristic vector decomposed by the matrix as the input of the multilayer perceptron, processing the potential characteristic vector by the multilayer perceptron to obtain a result predicted by a multilayer perceptron model, and entering the step S4;
step S4; and combining the result predicted by the multilayer perceptron model with the result of matrix decomposition in the merging layer, and optimizing the model by using a normalized cross entropy method to obtain the final predicted score.
As a preferred technical scheme of the invention: in the step S1, firstly, the LDA model is used to mine the hidden information of the relevant types in the user video review, generate the user type potential feature matrix LU and the video type potential feature matrix LV, and combine the user type potential feature matrix with the video type potential feature matrix to obtain the user type potential feature matrix LU and the video type potential feature matrix LV
To the hidden information matrix L, the calculation formula is:
Figure SMS_9
equation 4
And combining the hidden information matrix L with the original user-video scoring matrix R to generate a new user-video scoring matrix
Figure SMS_10
The calculation formula is as follows:
Figure SMS_11
equation 5
As a preferred technical scheme of the invention: in step S2, users with the same interests and hobbies may affect each other, and the user similarity may be calculated using a weighted pearson correlation coefficient:
Figure SMS_12
equation 6
In the formula 6, the first and second groups,
Figure SMS_13
and &>
Figure SMS_14
Represents the mean rating, based on the user u and v, respectively>
Figure SMS_15
Represents the rating of user u on video i, </or>
Figure SMS_16
Represents the rating of user v on video i, <' > or>
Figure SMS_17
Represents a set of videos that user u has commented on, and @>
Figure SMS_18
Represents a set of videos that user v has commented on, <' > or>
Figure SMS_19
The Jaccard correlation coefficient of the items which can influence the user similarity calculation is the weight, and the calculation formula is as follows:
Figure SMS_20
equation 7
Wherein,
Figure SMS_21
represents a set of videos that user u has commented on, and @>
Figure SMS_22
Representing a video set which is commented by a user v;
the user's score for an item depends on the influence of neighboring users and may also be influenced by neighboring users of neighboring users, but neighboring users at a certain distance do not influence the user, i.e. become unreliable, so a reliable value is introduced, and neighboring users with a reliable value greater than a certain value have an influence on the user item score, and the reliable value is calculated by:
Figure SMS_23
equation 8
In the formula 8, the process is described,
Figure SMS_26
represents a score of u on video i, <' > or>
Figure SMS_28
Represents the rating of user v on video i, <' > or>
Figure SMS_30
Represents a set of videos that user u has commented on, and @>
Figure SMS_25
Represents a set of videos that user v has commented on, <' > or>
Figure SMS_27
Represents the maximum value of the score, based on the evaluation>
Figure SMS_29
Indicates a trust distance, i.e. the number of users present between user u and user v, is present>
Figure SMS_31
Represents the maximum distance allowed between two users, and>
Figure SMS_24
is a correction parameter, is a number greater than 0 and less than 1, and is reliable
Figure SMS_32
Equation 9
Introducing a reliable nearest neighbor regular term to constrain the learning of the potential feature matrix of the user during matrix decomposition, wherein a reliable nearest neighbor regular term constraint function is as follows:
Figure SMS_33
. Equation 10
Wherein,
Figure SMS_34
is a potential vector for user u, <' > is>
Figure SMS_35
Is a potential feature vector for user v.
As a preferred technical scheme of the invention: in the step S3, the user potential feature vector and the video potential feature vector are used as inputs of a multilayer perceptron, wherein the deep neural network is composed of the multilayer perceptron and a single-layer perceptron, the multilayer perceptron includes an input layer, a plurality of hidden layers allowing neural structure nonlinearity and an output layer, and a result of a multilayer perceptron model is obtained by processing the multilayer perceptron through nonlinear features of the hidden layers.
As a preferred technical scheme of the invention: in step S4, the single-layer perceptron in the deep neural network structure is a merging layer, and the prediction result of the multilayer perceptron model is combined with the result of the regularized dual matrix decomposition model in the merging layer, where the calculation formula is:
Figure SMS_36
equation 11
In the formula 11
Figure SMS_37
For an activation function, <' >>
Figure SMS_38
For the set of matrix weights between the output layer and the merging layer, <' >>
Figure SMS_39
Is the result of the output layer, is>
Figure SMS_40
For a user potential vector, based on the number of potential vectors in the selected candidate set>
Figure SMS_41
For a video potential vector>
Figure SMS_42
And optimizing the model by using a normalized cross entropy method for merging the deviation items of the layers to finally obtain the prediction score.
Has the beneficial effects that:
1. the method introduces the video association regular term and the reliable nearest neighbor regular term which are blended into the user activity to restrain the learning of the potential feature matrix, utilizes the nonlinear structure of the deep neural network to relieve the limitation of linear dot products in the matrix decomposition process, combines the result of the deep neural network model with the result of the double regularization matrix decomposition, and improves the precision of video scoring prediction.
2. According to the method, the LDA model is utilized to mine relevant information in the user video comments, a user type potential feature matrix and a video type potential feature matrix are generated, the user type potential feature matrix and the video type potential feature matrix are combined to obtain a hidden information matrix, the hidden information matrix and an original user video scoring matrix are combined to generate a new user-video scoring matrix, and the problems of cold start and data sparsity are solved.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a block diagram of the multi-level perceptron of the present invention;
FIG. 3 is a diagram of the deep neural network architecture of the present invention.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The method is characterized in that a user-video scoring matrix is reconstructed on the basis of a traditional matrix decomposition model, and a video association regular term and a reliable nearest neighbor regular term which are blended into the user activity are introduced to limit the learning of a potential feature matrix. The nonlinear structure of the deep neural network is utilized, the limit of linear dot products in the matrix decomposition process is relieved, the potential feature vector decomposed by the matrix is used as the input of the deep neural network, the result of the MLP model is obtained through the processing of a plurality of layers of perceptrons, the result of the MLP model is combined with the result of the biregularized matrix decomposition model in a single-layer perceptron layer, namely a merging layer, the model is optimized through a normalized cross entropy method, and therefore the scoring prediction precision is improved.
As shown in fig. 1.
The invention designs a video scoring prediction method based on a deep neural network and double regularization, which is used for improving the precision of the scoring prediction of a recommended video, and comprises the following steps:
step S1: processing the video comments, excavating hidden information, combining a hidden information matrix with an original user-video scoring matrix to generate a new user-video scoring matrix, and entering the step S2;
step S2: adding a double regular term for restricting the learning of the potential feature matrix when decomposing the user-video scoring matrix, and entering a step S3;
and step S3: taking the potential characteristic vector decomposed by the matrix as the input of the multilayer perceptron, processing the potential characteristic vector by the multilayer perceptron to obtain a result predicted by a multilayer perceptron model, and entering the step S4;
and step S4: and combining the result predicted by the multilayer perceptron model with the result of matrix decomposition in the merging layer, and optimizing the model by using a normalized cross entropy method to obtain the final predicted score.
The method comprises the following specific steps:
the step S1 comprises the following steps: firstly, mining related types of hidden information in user video review by using an LDA (latent Dirichlet Allocation) model to generate a user type potential feature matrix LU and a video type potential feature matrix LV, and combining the user type potential feature matrix with the video type potential feature matrix to obtain a hidden information matrix L, wherein the calculation formula is as follows
Figure SMS_43
And combining the hidden information matrix L with the original user-video scoring matrix R to reconstruct a user-video scoring matrix R, wherein the calculation formula is as follows:
Figure SMS_44
。
The step S2 comprises the following steps: carrying out matrix decomposition on the user-video scoring matrix, and decomposing the high-dimensional user-video scoring matrix into a low-dimensional user characteristic matrix and a video characteristic matrix, wherein the formula is as follows:
Figure SMS_45
wherein U represents a user feature matrix, U i Potential feature vector representing user i, V represents a video feature matrix, V j Is the potential feature vector for video j. The low-dimensional matrix decomposition method approximately calculates the scoring matrix R by the product of the d-rank factors. The prediction score of user i for video j is expressed as
Figure SMS_46
The square of the error between the predicted score and the raw score is taken as a loss function, which is minimized to approximate the score matrix R. The loss function is:
Figure SMS_47
，
in the above-mentioned formula,
Figure SMS_48
is an indicator function that indicates that if user i scores item j, it equals 1, otherwise it equals 0.
Figure SMS_49
And &>
Figure SMS_50
Over-fitting is prevented for two regular terms. Since each user scores the video to make a certain contribution to the video similarity, but each user contribution is not exactly the same, and the video similarity can be divided into active users and inactive users by considering the activity of the users, the active users refer to users who have a large number of score records on the video, and the inactive users refer to users who only have score records on a small number of videos, so the contributions of the active users and the inactive users should be distinguished when calculating the video similarity, and the activity of the users can be defined as:
Figure SMS_51
in the above-mentioned formula,
Figure SMS_52
representing the total number of scores for user u. Therefore, the video similarity calculation method obtained by combining the liveness coefficient of the user with the modified cosine similarity is as follows:
Figure SMS_53
in the above-mentioned formula,
Figure SMS_54
represents the rating of user u on video i, </or>
Figure SMS_55
Represents the rating of user u on video j, and>
Figure SMS_56
representing the user u's score. The video association regularization item integrated with the user activity is introduced during matrix decomposition to constrain the learning of a project potential feature matrix, and the video association regularization constraint function formula integrated with the user activity is as follows:
Figure SMS_57
wherein V represents a video feature matrix, V j Is a potential feature vector, V, of video j i Are potential feature vectors for video i. Users with the same interest will influence each other, and the user similarity can be calculated using weighted pearson correlation coefficients:
Figure SMS_58
in the above-mentioned formula,
Figure SMS_59
and &>
Figure SMS_60
Represents the mean score of users u and v, respectively>
Figure SMS_61
The Jaccard correlation coefficient of the items which can influence the user similarity calculation is the weight, and the calculation formula is as follows:
Figure SMS_62
wherein,
Figure SMS_63
represents a set of videos that user u has commented on, and @>
Figure SMS_64
Representing a set of videos that user v has commented on.
The user's score for an item depends on the influence of neighboring users and may also be influenced by neighboring users of neighboring users, but neighboring users at a certain distance do not influence the user, i.e. become unreliable, so a reliable value is introduced, and neighboring users with a reliable value greater than a certain value have an influence on the user item score, and the reliable value is calculated by:
Figure SMS_65
，
in the above-mentioned formula,
Figure SMS_67
represents a score of u on video i, <' > or>
Figure SMS_69
Represents the rating of user v on video i, <' > or>
Figure SMS_72
Represents a set of videos that user u has commented on, and @>
Figure SMS_68
Represents a set of videos that a user v has commented on,/>>
Figure SMS_70
Represents the maximum value of the score, based on the evaluation>
Figure SMS_71
Representing a trust distance, i.e. the number of users present between user u and user v, in conjunction with a user v>
Figure SMS_73
Represents the maximum distance allowed between two users, and>
Figure SMS_66
is a number greater than 0 and less than 1 for the correction parameter. Reliable nearest neighbor users are
Figure SMS_74
，
Introducing a reliable nearest neighbor regular term to constrain the learning of a potential feature matrix of a user during matrix decomposition, wherein a reliable nearest neighbor regular term constraint function is as follows:
Figure SMS_75
wherein,
Figure SMS_76
is a potential vector for user u, <' > is>
Figure SMS_77
Is a potential feature vector for user v.
Adding a video associated regularization term fused with user activity and a reliable nearest neighbor regularization term, wherein the final optimization loss function is as follows:
Figure SMS_78
and searching an optimal solution by adopting a random gradient descent method, and finding out an optimal potential feature matrix.
As shown in fig. 2.
The step S3 comprises the following steps: user potential feature vector U u And video latent feature vector V i As input to a multi-layered perceptron, as shown in FIG. 2, comprising an input layer L in A plurality of hidden layers allowing the nonlinearity of the neural structure and an output layer L out An input layer L in The output vector of (a) is:
Figure SMS_79
outputting a vector after the first hidden layer processing:
Figure SMS_80
wherein
Figure SMS_81
Is a set of weights, included in a matrix between the input layer and the first layer hidden layer L1, is->
Figure SMS_82
Is a aberration of level L1>
Figure SMS_83
Is an activation function of
Figure SMS_84
So as to hide the layer L k The output vector of (a) is:
Figure SMS_85
wherein
Figure SMS_86
Is an activation function of the neuron>
Figure SMS_87
Is a weight matrix, is based on>
Figure SMS_88
Is a deviation. Multilayer perceptron output layer L out The output vector is:
Figure SMS_89
as shown in fig. 3.
The step S4 comprises the following steps: the single-layer perceptron in the deep neural network structure is a merging layer, wherein the deep neural network structure is shown in fig. 3, the prediction result of the multilayer perceptron model and the result of the double regularization matrix decomposition model are combined in the merging layer, and the calculation formula is as follows:
Figure SMS_90
wherein,
Figure SMS_91
for an activation function, <' >>
Figure SMS_92
For the set of matrix weights between the output layer and the merging layer, <' >>
Figure SMS_93
Is the result of the output layer, is>
Figure SMS_94
For a user potential vector, based on the number of potential vectors in the selected candidate set>
Figure SMS_95
For a potential vector of video, be>
Figure SMS_96
To merge the deviation terms of the layers
The proposed model is continuously optimized by using a normalized cross entropy method through the following cost functions:
Figure SMS_97
in the above-mentioned formula, the first and second,
Figure SMS_98
for merging layer neuron number, based on the combined layer neuron number>
Figure SMS_99
For a predicted score>
Figure SMS_100
For a real point of training instance, <' > based on the number of true points>
Figure SMS_101
The maximum value of the score is indicated. The cost function continuously optimizes the model by using a gradient descent method to obtain the final prediction score.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.

Claims (5)

1. A video scoring prediction method based on a deep neural network and double regularization is characterized by comprising the following steps:
step S1: processing the video comments, excavating hidden information, combining a hidden information matrix with an original user-video scoring matrix to generate a new user-video scoring matrix, and entering the step S2;
step S2: the learning of a double regular term constraint potential feature matrix is added during the decomposition of a user-video scoring matrix, the scoring of each user on a video makes a certain contribution to the video similarity, the user contributions are not completely the same, the video scoring matrix can be divided into active users and inactive users according to the liveness of the users, the active users refer to users who record a large number of scoring records on the video, the inactive users refer to users who record only a small number of scoring records on the video, so the contributions of the active users and the inactive users are separated when the video similarity is calculated, and the liveness of the users is defined as:
Figure QLYQS_1
equation 1
In the formula 1, the first and second groups of the compound,
Figure QLYQS_2
the total score of the user u is represented, so that the video similarity calculation method obtained by combining the activity coefficient of the user and the modified cosine similarity comprises the following steps:
Figure QLYQS_3
equation 2
In the case of the formula 2, the,
Figure QLYQS_4
represents the rating of user u on video i, </or>
Figure QLYQS_5
Represents the rating of user u on video j, and>
Figure QLYQS_6
represents a rating score for user u, and->
Figure QLYQS_7
Representing a set of users who have simultaneously scored videos i and j; and (3) introducing the learning of the video associated regularization term constraint project potential feature matrix blended with the user activity during matrix decomposition, wherein the video associated regularization constraint function formula blended with the user activity at the moment is as follows:
Figure QLYQS_8
equation 3
In equation 3, V represents the video feature matrix, V j Is a potential feature vector, V, of video j i Is a potential feature vector of video i and proceeds to step S3;
and step S3: taking the potential characteristic vector decomposed by the matrix as the input of the multilayer perceptron, processing the potential characteristic vector by the multilayer perceptron to obtain a result predicted by a multilayer perceptron model, and entering the step S4;
and step S4: and combining the result predicted by the multilayer perceptron model with the result of matrix decomposition in the merging layer, and optimizing the model by using a normalized cross entropy method to obtain the final predicted score.
2. The method of claim 1, wherein the video score prediction method based on deep neural network and regularization comprises: in the step S1, firstly, the LDA model is used to mine the hidden information of the relevant type in the user video review, the user type potential feature matrix LU and the video type potential feature matrix LV are generated, and the user type potential feature matrix is combined with the video type potential feature matrix to obtain the hidden information matrix L, wherein the calculation formula is as follows:
Figure QLYQS_9
equation 4
And combining the hidden information matrix L with the original user-video scoring matrix R to generate a new user-video scoring matrix
Figure QLYQS_10
The calculation formula is as follows:
Figure QLYQS_11
equation 5.
3. The method of claim 1, wherein the video score prediction method based on deep neural network and regularization comprises: in step S2, users with the same interests and hobbies affect each other, and the user similarity is calculated using the weighted pearson correlation coefficient:
Figure QLYQS_12
equation 6
In the formula 6, the first and second groups,
Figure QLYQS_13
and &>
Figure QLYQS_14
Represents the mean rating, based on the user u and v, respectively>
Figure QLYQS_15
Represents the rating of user u on video i, </or>
Figure QLYQS_16
Represents the rating of user v on video i, <' > or>
Figure QLYQS_17
Represents a set of videos that user u has commented on, and @>
Figure QLYQS_18
Represents a set of videos that user v has commented on,
Figure QLYQS_19
the Jaccard correlation coefficient of the items which can influence the user similarity calculation is the weight, and the calculation formula is as follows:
Figure QLYQS_20
equation 7
Wherein,
Figure QLYQS_21
represents a set of videos that user u has commented on, and @>
Figure QLYQS_22
Representing a video set which is commented by a user v;
the user's score for an item depends on the influence of neighboring users and is also influenced by neighboring users of neighboring users, but neighboring users at a certain distance do not influence the user, i.e. become unreliable, so a reliable value is introduced, and neighboring users with a reliable value greater than a certain value have an influence on the user item score, and the reliable value is calculated by:
Figure QLYQS_23
equation 8
In the formula 8, the process is described,
Figure QLYQS_26
represents a score of u on video i, <' > or>
Figure QLYQS_27
Represents the rating on video i of user v>
Figure QLYQS_29
Represents a set of videos that user u has commented on, and @>
Figure QLYQS_25
Represents a set of videos that user v has commented on, <' > or>
Figure QLYQS_28
Represents the maximum value of the score, based on the evaluation>
Figure QLYQS_30
Representing a trust distance, i.e. the number of users present between user u and user v, in conjunction with a user v>
Figure QLYQS_31
Represents the maximum distance allowed between two users, and>
Figure QLYQS_24
is to correct the parameter, is oneA number greater than 0 and less than 1, the reliable nearest neighbor user is
Figure QLYQS_32
Equation 9
Introducing a reliable nearest neighbor regular term to constrain the learning of the potential feature matrix of the user during matrix decomposition, wherein a reliable nearest neighbor regular term constraint function is as follows:
Figure QLYQS_33
equation 10
Wherein,
Figure QLYQS_34
for a potential vector of user u>
Figure QLYQS_35
Is a potential feature vector for user v.
4. The method of claim 1, wherein the video score prediction method based on deep neural network and regularization comprises: in the step S3, the user potential feature vector and the video potential feature vector are used as inputs of a multilayer perceptron, wherein the deep neural network is composed of the multilayer perceptron and a single-layer perceptron, the multilayer perceptron includes an input layer, a plurality of hidden layers allowing the neural structure to be nonlinear, and an output layer, and the result of the multilayer perceptron model is obtained through processing by the multilayer perceptron by utilizing the nonlinear features of the hidden layers.
5. The method of claim 1, wherein the video score prediction method based on deep neural network and regularization comprises: in step S4, the single-layer perceptron in the deep neural network structure is a merging layer, and the prediction result of the multilayer perceptron model is combined with the result of the regularized dual matrix decomposition model in the merging layer, where the calculation formula is:
Figure QLYQS_36
equation 11
In equation 11
Figure QLYQS_37
For an activation function, <' >>
Figure QLYQS_38
For the set of matrix weights between the output layer and the merging layer, <' >>
Figure QLYQS_39
Is the result of the output layer, is>
Figure QLYQS_40
For a user potential vector>
Figure QLYQS_41
For a potential vector of video, be>
Figure QLYQS_42
And optimizing the model by using a normalized cross entropy method for merging the deviation items of the layers to finally obtain the prediction score. />
CN202310187456.2A 2023-03-02 2023-03-02 Video scoring prediction method based on deep neural network and double regularization Active CN115905617B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202310187456.2A CN115905617B (en) 2023-03-02 2023-03-02 Video scoring prediction method based on deep neural network and double regularization
PCT/CN2023/135369 WO2024179089A1 (en) 2023-03-02 2023-11-30 Video score prediction method based on deep neural network and double regularization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310187456.2A CN115905617B (en) 2023-03-02 2023-03-02 Video scoring prediction method based on deep neural network and double regularization

Publications (2)

Publication Number Publication Date
CN115905617A true CN115905617A (en) 2023-04-04
CN115905617B CN115905617B (en) 2023-05-30

Family

ID=85744775

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310187456.2A Active CN115905617B (en) 2023-03-02 2023-03-02 Video scoring prediction method based on deep neural network and double regularization

Country Status (2)

Country Link
CN (1) CN115905617B (en)
WO (1) WO2024179089A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024179089A1 (en) * 2023-03-02 2024-09-06 南京邮电大学 Video score prediction method based on deep neural network and double regularization

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112328900A (en) * 2020-11-27 2021-02-05 北京工业大学 Deep learning recommendation method integrating scoring matrix and comment text
CN112668374A (en) * 2019-10-16 2021-04-16 北京灵汐科技有限公司 Image processing method and device, re-recognition network training method and electronic equipment
CN113157957A (en) * 2021-03-05 2021-07-23 北京工业大学 Attribute graph document clustering method based on graph convolution neural network
CN113807422A (en) * 2021-09-07 2021-12-17 南京邮电大学 Weighted graph convolutional neural network score prediction model fusing multi-feature information

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111581333B (en) * 2020-05-07 2023-05-26 重庆大学 Text-CNN-based video and audio play list pushing method and video and audio play list pushing system
CN114117191A (en) * 2020-08-26 2022-03-01 付昳漫 Context-aware recommendation method based on deep neural network
CN114625980A (en) * 2020-12-13 2022-06-14 国网吉林省电力有限公司培训中心 Dual-regularization collaborative filtering recommendation method
CN115905617B (en) * 2023-03-02 2023-05-30 南京邮电大学 Video scoring prediction method based on deep neural network and double regularization

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112668374A (en) * 2019-10-16 2021-04-16 北京灵汐科技有限公司 Image processing method and device, re-recognition network training method and electronic equipment
CN112328900A (en) * 2020-11-27 2021-02-05 北京工业大学 Deep learning recommendation method integrating scoring matrix and comment text
CN113157957A (en) * 2021-03-05 2021-07-23 北京工业大学 Attribute graph document clustering method based on graph convolution neural network
CN113807422A (en) * 2021-09-07 2021-12-17 南京邮电大学 Weighted graph convolutional neural network score prediction model fusing multi-feature information

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024179089A1 (en) * 2023-03-02 2024-09-06 南京邮电大学 Video score prediction method based on deep neural network and double regularization

Also Published As

Publication number Publication date
WO2024179089A1 (en) 2024-09-06
CN115905617B (en) 2023-05-30

Similar Documents

Publication Publication Date Title
CN110929164B (en) Point-of-interest recommendation method based on user dynamic preference and attention mechanism
Pan et al. A correlative denoising autoencoder to model social influence for top-N recommender system
CN111127142B (en) Article recommendation method based on generalized nerve attention
Ni et al. An effective recommendation model based on deep representation learning
Ni et al. A two-stage embedding model for recommendation with multimodal auxiliary information
CN110866145B (en) Co-preference-assisted deep single-class collaborative filtering recommendation method
CN107507073A (en) Based on the service recommendation method for trusting extension and the sequence study of list level
CN112800344B (en) Deep neural network-based movie recommendation method
CN111125540A (en) Recommendation method integrating knowledge graph representation learning and bias matrix decomposition
CN115905617B (en) Video scoring prediction method based on deep neural network and double regularization
Lin et al. Lifelong Text-Audio Sentiment Analysis learning
Lin et al. Transfer learning for collaborative recommendation with biased and unbiased data
CN113836393B (en) Cold start recommendation method based on preference self-adaptive meta-learning
CN110889505A (en) Cross-media comprehensive reasoning method and system for matching image-text sequences
Nazari et al. Scalable and data-independent multi-agent recommender system using social networks analysis
Mu et al. Auxiliary stacked denoising autoencoder based collaborative filtering recommendation
Rauf et al. BCE4ZSR: Bi-encoder empowered by teacher cross-encoder for zero-shot cold-start news recommendation
CN115310004A (en) Graph nerve collaborative filtering recommendation method fusing project time sequence relation
CN116049377A (en) Context-aware recommendation system and method based on matrix decomposition and CRNN
Xiao et al. A better understanding of the interaction between users and items by knowledge graph learning for temporal recommendation
Lu et al. GTPAN: Global Target Preference Attention Network for session-based recommendation
Dashti et al. Hybrid recommender system based on fuzzy neural algorithm
Gan et al. A WEIGHTED NEURAL MATRIX FACTORIZATION HEALTH MANAGEMENT RECOMMENDATION ALGORITHM INTEGSCORING DEEP LEARNING TECHNOLOGY
AL SBOU et al. Performance comparison of three different types of autoencoders using recommendation systems
Tomar et al. Twin support vector machine for multiple instance learning based on bag dissimilarities

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