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

CN115905617B - 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
CN115905617B
CN115905617B CN202310187456.2A CN202310187456A CN115905617B CN 115905617 B CN115905617 B CN 115905617B CN 202310187456 A CN202310187456 A CN 202310187456A CN 115905617 B CN115905617 B CN 115905617B
Authority
CN
China
Prior art keywords
user
video
matrix
rating
users
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
CN202310187456.2A
Other languages
Chinese (zh)
Other versions
CN115905617A (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

本发明涉及一种基于深度神经网络与双正则化的视频评分预测方法，重构用户‑视频评分矩阵，引入融有用户活跃度的视频关联正则项和可靠的最近邻正则项，构建了融合融有用户活跃度的视频关联正则项和可靠的最近邻正则项的矩阵分解推荐模型，将潜在特征输入深度神经网络中得到深度神经网络模型的结果，并将深度神经网络模型的结果与矩阵分解的结构结合得到最终的预测评分，提高了预测评分的精度；并利用LDA模型挖掘用户视频评论中相关信息，生成用户类型潜在特征矩阵和视频类型潜在特征矩阵，并将两者结合得到隐藏的信息矩阵，再将隐藏的信息矩阵与原始用户视频评分矩阵相结合生成新的用户‑视频评分矩阵，缓解了冷启动和数据稀疏性问题。

Figure 202310187456

The present invention relates to a video scoring prediction method based on deep neural network and double regularization, which reconstructs the user-video scoring matrix, introduces video association regularization items incorporating user activity and reliable nearest neighbor regularization items, and constructs a fusion The matrix factorization recommendation model of the video association regularization item with user activity and the reliable nearest neighbor regularization item, the latent features are input into the deep neural network to obtain the result of the deep neural network model, and the result of the deep neural network model is compared with the matrix factorization Combining the structure to get the final prediction score improves the accuracy of the prediction score; and using the LDA model to mine relevant information in user video comments, generate user type latent feature matrix and video type latent feature matrix, and combine the two to get the hidden information matrix , and then combine the hidden information matrix with the original user video rating matrix to generate a new user-video rating matrix, which alleviates the cold start and data sparsity problems.

Figure 202310187456

Description

一种基于深度神经网络与双正则化的视频评分预测方法A video rating prediction method based on deep neural network and double regularization

技术领域Technical Field

本发明涉及一种基于深度神经网络与双正则化的视频评分预测方法，属于评分预测领域。The invention relates to a video rating prediction method based on a deep neural network and double regularization, and belongs to the field of rating prediction.

背景技术Background Art

随着互联网技术的快速发展，各个网络平台中视频资源越来越多，为用户提供丰富的视频资源，为用户提供更多的选择时，也给我们带来了麻烦和困扰，庞大的视频资源不仅增加了我们寻找自己喜好视频的难度，而且使得寻找视频的过程变得十分耗费时间。为了解决信息过载问题，个性化推荐系统成为解决这一个问题的有效工具。评分预测又是推荐算法的重要组成部分。现有的推荐算法主要有三大类：基于协同过滤的推荐算法、基于内容的推荐算法和混合推荐算法。目前使用最多的是基于协同过滤的推荐算法，而协同过滤推荐算法中使用最多的是基于模型的协同过滤推荐算法，基于模型的协同过滤推荐算法中比较常见的几种算法包括：矩阵分解模型、奇异值分解、聚类分析等。但是现存的协同过滤推荐算法存在数据稀疏及冷启动等问题，导致对推荐的视频资源评分预测不准确，从而影响个性化推荐的结果，如何提高对视频资源的预测评分的准确性，进一步提高推荐精度成为当前研究的热点之一。With the rapid development of Internet technology, there are more and more video resources in various network platforms. While providing users with rich video resources and more choices, it also brings us troubles and troubles. The huge video resources not only increase the difficulty of finding videos we like, but also make the process of finding videos very time-consuming. In order to solve the problem of information overload, personalized recommendation system has become an effective tool to solve this problem. Rating prediction is an important part of recommendation algorithm. There are three main categories of existing recommendation algorithms: recommendation algorithm based on collaborative filtering, recommendation algorithm based on content and hybrid recommendation algorithm. At present, the most used recommendation algorithm is based on collaborative filtering, and the most used in collaborative filtering recommendation algorithm is model-based collaborative filtering recommendation algorithm. The more common algorithms in 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 such as data sparsity and cold start, which leads to inaccurate prediction of the recommended video resource rating, thus affecting the results of personalized recommendation. How to improve the accuracy of the predicted rating of video resources and further improve the recommendation accuracy has become one of the current research hotspots.

发明内容Summary of the invention

本发明的目的在于针对上述现有技术的不足，提供了一种基于深度神经网络与双正则化的视频评分预测方法，重构用户-视频评分矩阵，在矩阵分解时引入融入用户活跃度的视频关联正则项和可靠的最近邻正则项来约束潜在特征矩阵的学习，并引入深度神经网络，利用深度神经网络的非线性特征减轻矩阵分解过程中的线性点积的限制，将深度神经网络模型的结果与双正则化矩阵分解的结果相结合，提高视频评分预测的精度。The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and provide a video rating prediction method based on deep neural network and double regularization, reconstruct the user-video rating matrix, introduce video association regularization terms that incorporate user activity and reliable nearest neighbor regularization terms during matrix decomposition to constrain the learning of the latent feature matrix, introduce deep neural network, utilize the nonlinear characteristics of deep neural network to alleviate the limitation of linear dot product in the matrix decomposition process, combine the results of deep neural network model with the results of double regularization matrix decomposition, and improve the accuracy of video rating prediction.

本发明采用的技术方案如下：一种基于深度神经网络与双正则化的视频评分预测方法，用于提高对推荐视频评分预测的精度，具体包括如下步骤：The technical solution adopted by the present invention is as follows: a video rating prediction method based on deep neural network and double regularization, which is used to improve the accuracy of predicting the rating of recommended videos, and specifically includes the following steps:

步骤S1：对视频评论进行处理，挖掘出隐藏信息，并将隐藏信息矩阵与原始用户-视频评分矩阵相结合生成新的用户-视频评分矩阵，并进入步骤S2；Step S1: Process the video comments, mine hidden information, and combine the hidden information matrix with the original user-video rating matrix to generate a new user-video rating matrix, and then proceed to step S2;

步骤S2：对用户-视频评分矩阵分解时加入双正则项约束潜在特征矩阵的学习，每个用户对视频的评分都会对视频相似度做出一定的贡献，用户贡献却不完全相同，从用户的活跃度考虑能够分为活跃用户和不活跃用户，活跃用户指的是对视频有大量的评分记录的用户，而不活跃用户是指只对少数视频进行评分记录的用户，所以在计算视频相似度时应将活跃用户和不活跃用户的贡献区分开，用户的活跃度定义为：Step S2: When decomposing the user-video rating matrix, double regularization terms are added to constrain the learning of the latent feature matrix. Each user's rating of a video will make a certain contribution to the video similarity, but the user contributions are not exactly the same. From the perspective of user activity, they can be divided into active users and inactive users. Active users refer to users who have a large number of rating records for videos, while inactive users refer to users who have only rated a few videos. Therefore, when calculating video similarity, the contributions of active users and inactive users should be distinguished. The user's activity is defined as:

Figure SMS_1
公式1
Figure SMS_1
Formula 1

在公式1中，

Figure SMS_2
表示用户u的评分总量，因此用户的活跃度系数结合修正的余弦相似度得到的视频相似度计算方法为：In formula 1,
Figure SMS_2
represents the total number of ratings of user u, so the video similarity calculation method obtained by combining the user's activity coefficient with the modified cosine similarity is:

Figure SMS_3
公式2
Figure SMS_3
Formula 2

在公式2中，

Figure SMS_4
表示用户u对视频i的评分，
Figure SMS_5
表示用户u对视频j的评分，
Figure SMS_6
表示用户u的评分评分，
Figure SMS_7
表示同时对视频i和j有过评分的用户集合；在矩阵分解时引入融入用户活跃度的视频关联正则项约束项目潜在特征矩阵的学习，此时融入用户活跃度的视频关联正则化约束函数公式为：In formula 2,
Figure SMS_4
represents the rating of video i by user u,
Figure SMS_5
represents the rating of video j by user u,
Figure SMS_6
represents the rating of user u,
Figure SMS_7
represents the set of users who have rated videos i and j at the same time; when decomposing the matrix, the video association regularization term that incorporates user activity is introduced to constrain the learning of the project's latent feature matrix. At this time, the formula of the video association regularization constraint function that incorporates user activity is:

Figure SMS_8
公式3
Figure SMS_8
Formula 3

在公式3中，V表示视频特征矩阵，Vj是视频j的潜在特征向量，Vi是视频i的潜在特征向量，并进入步骤S3；In Formula 3, V represents the video feature matrix, Vj is the potential feature vector of video j, Vi is the potential feature vector of video i, and the process proceeds to step S3;

步骤S3：将矩阵分解出的潜在特征向量作为多层感知机的输入，经过多层感知机处理得到多层感知机模型预测的结果，并进入步骤S4；Step S3: The potential feature vector decomposed from the matrix is used as the input of the multi-layer perceptron, and the result predicted by the multi-layer perceptron model is obtained after being processed by the multi-layer perceptron, and then the process goes to step S4;

步骤S4；在合并层将多层感知机模型预测的结果与矩阵分解的结果相结合，使用归一化交叉熵法来优化模型，得到最终预测得分。Step S4: Combine the prediction results of the multilayer perceptron model with the results of matrix decomposition at the merging layer, use the normalized cross entropy method to optimize the model, and obtain the final prediction score.

作为本发明的一种优选技术方案：所述步骤S1中，首先利用LDA模型挖掘用户视频评论中相关类型的隐藏信息，生成用户类型潜在特征矩阵LU和视频类型潜在特征矩阵LV，并将用户类型潜在特征矩阵与视频类型潜在特征矩阵相结合得As a preferred technical solution of the present invention: in the step S1, the LDA model is first used to mine the hidden information of related types in the user video comments, 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

到隐藏信息矩阵L，其计算公式为：To the hidden information matrix L, its calculation formula is:

Figure SMS_9
公式4
Figure SMS_9
Formula 4

并将隐藏信息矩阵L与原始用户-视频评分矩阵R相结合生成新的用户-视频评分矩阵

Figure SMS_10
，其计算公式为：The hidden information matrix L is combined with the original user-video rating matrix R to generate a new user-video rating matrix
Figure SMS_10
, and its calculation formula is:

Figure SMS_11
公式5
Figure SMS_11
Formula 5

作为本发明的一种优选技术方案：所述步骤S2中，相同兴趣爱好的用户会互相影响，用户相似度可以使用加权的皮尔逊相关系数进行计算：As a preferred technical solution of the present invention: in step S2, users with the same interests and hobbies will influence each other, and the user similarity can be calculated using the weighted Pearson correlation coefficient:

Figure SMS_12
公式6
Figure SMS_12
Formula 6

在公式6中，

Figure SMS_13
和
Figure SMS_14
分别代表用户u和v的平均评分，
Figure SMS_15
表示用户u对视频i的评分，
Figure SMS_16
表示用户v对视频i的评分，
Figure SMS_17
表示用户u评论过的视频集合，
Figure SMS_18
表示用户v评论过的视频集合，
Figure SMS_19
为权重即能够影响用户相似度计算的项目的Jaccard相关系数，其计算公式如下：In Formula 6,
Figure SMS_13
and
Figure SMS_14
Represent the average ratings of users u and v respectively,
Figure SMS_15
represents the rating of video i by user u,
Figure SMS_16
represents the rating of user v on video i,
Figure SMS_17
represents the set of videos commented by user u,
Figure SMS_18
represents the set of videos commented by user v,
Figure SMS_19
The Jaccard correlation coefficient of the item that can affect the calculation of user similarity is the weight, and its calculation formula is as follows:

Figure SMS_20
公式7
Figure SMS_20
Formula 7

其中，

Figure SMS_21
表示用户u评论过的视频集合，
Figure SMS_22
表示用户v评论过的视频集合；in,
Figure SMS_21
represents the set of videos commented by user u,
Figure SMS_22
represents the set of videos commented by user v;

用户对项目的评分取决于邻近用户的影响，也可能受到邻近用户的邻近用户的影响，但是一定距离后的邻近用户不会在对用户有所影响即变得不在可靠，因此引进可靠值，可靠值大于一定值的邻近用户会对用户项目评分有影响，可靠值计算方式为：The user's rating of an item depends on the influence of neighboring users, and may also be affected by the neighboring users of the neighboring users. However, neighboring users beyond a certain distance will no longer have an impact on the user and become unreliable. Therefore, a reliability value is introduced. Neighboring users with a reliability value greater than a certain value will have an impact on the user's item rating. The reliability value is calculated as follows:

Figure SMS_23
公式8
Figure SMS_23
Formula 8

在公式8中，

Figure SMS_26
表示用u对视频i的评分，
Figure SMS_28
表示用户v对视频i的评分，
Figure SMS_30
表示用户u评论过的视频集合，
Figure SMS_25
表示用户v评论过的视频集合，
Figure SMS_27
表示评分的最大值，
Figure SMS_29
表示信任距离即用户u和用户v之间存在的用户人数，
Figure SMS_31
表示两个用户之间允许的最大距离，
Figure SMS_24
是为修正参数，是一个大于0小于1的数，可靠的最近邻用户为In formula 8,
Figure SMS_26
represents the rating of video i by u,
Figure SMS_28
represents the rating of user v on video i,
Figure SMS_30
represents the set of videos commented by user u,
Figure SMS_25
represents the set of videos commented by user v,
Figure SMS_27
represents the maximum value of the score,
Figure SMS_29
represents the trust distance, that is, the number of users between user u and user v.
Figure SMS_31
Indicates the maximum distance allowed between two users.
Figure SMS_24
is a correction parameter, which is a number greater than 0 and less than 1. The reliable nearest neighbor user is

Figure SMS_32
公式9
Figure SMS_32
Formula 9

在矩阵分解时引入可靠的最近邻正则项来约束用户潜在特征矩阵的学习，可靠的最近邻正则项约束函数为：During matrix decomposition, a reliable nearest neighbor regularization term is introduced to constrain the learning of the user's potential feature matrix. The constraint function of the reliable nearest neighbor regularization term is:

Figure SMS_33
。公式10
Figure SMS_33
Formula 10

其中，

Figure SMS_34
为用户u的潜在向量，
Figure SMS_35
为用户v的潜在特征向量。in,
Figure SMS_34
is the potential vector of user u,
Figure SMS_35
is the potential feature vector of user v.

作为本发明的一种优选技术方案：所述步骤S3中，将用户潜在特征向量与视频潜在特征向量作为多层感知机的输入，其中该深度神经网络由多层感知机、单层感知机组成，其中多层感知机包括输入层、若干个允许神经结构非线性的隐藏层和输出层，利用隐藏层的非线性特征，通过多层感知机处理得到多层感知机模型的结果。As a preferred technical solution of the present invention: in the step S3, the user potential feature vector and the video potential feature vector are used as inputs of a multi-layer perceptron, wherein the deep neural network is composed of a multi-layer perceptron and a single-layer perceptron, wherein the multi-layer perceptron includes an input layer, several hidden layers that allow the nonlinearity of the neural structure, and an output layer, and the result of the multi-layer perceptron model is obtained by processing the multi-layer perceptron using the nonlinear characteristics of the hidden layer.

作为本发明的一种优选技术方案：所述步骤S4中，在该深度神经网络结构中单层感知机为合并层，在合并层将多层感知机模型的预测结果与双正则化矩阵分解模型的结果相结合，其计算公式为：As a preferred technical solution of the present invention: in the step S4, the single-layer perceptron in the deep neural network structure is a merging layer, and the prediction results of the multi-layer perceptron model are combined with the results of the double regularization matrix decomposition model in the merging layer, and the calculation formula is:

Figure SMS_36
公式11
Figure SMS_36
Formula 11

在公式11中

Figure SMS_37
为激活函数，
Figure SMS_38
为输出层与合并层之间的矩阵权重集，
Figure SMS_39
为输出层的结果，
Figure SMS_40
为用户潜在向量，
Figure SMS_41
为视频潜在向量，
Figure SMS_42
为合并层的偏差项，使用归一化交叉熵法来优化模型，最终得到预测评分。In formula 11
Figure SMS_37
is the activation function,
Figure SMS_38
is the matrix weight set between the output layer and the merging layer,
Figure SMS_39
is the result of the output layer,
Figure SMS_40
is the user latent vector,
Figure SMS_41
is the video latent vector,
Figure SMS_42
The normalized cross entropy method is used to optimize the model for the bias term of the merging layer, and finally the predicted score is obtained.

有益效果：Beneficial effects:

1. 本发明引入融入用户活跃度的视频关联正则项和可靠的最近邻正则项来约束潜在特征矩阵的学习，利用深度神经网络的非线性结构来减轻矩阵分解过程中线性点积的限制，并将深度神经网络模型的结果与双正则化矩阵分解的结果相结合，提高了视频评分预测的精度。1. The present invention introduces a video association regularization term that incorporates user activity and a reliable nearest neighbor regularization term to constrain the learning of the latent feature matrix, uses the nonlinear structure of the deep neural network to alleviate the limitation of the linear dot product in the matrix decomposition process, and combines the results of the deep neural network model with the results of the double regularization matrix decomposition, thereby improving the accuracy of video score prediction.

2. 本发明利用LDA模型挖掘用户视频评论中相关信息，生成用户类型潜在特征矩阵和视频类型潜在特征矩阵，并将用户类型潜在特征矩阵与视频类型潜在特征矩阵相结合得到隐藏的信息矩阵，并将隐藏的信息矩阵与原始用户视频评分矩阵相结合生成新的用户-视频评分矩阵，缓解了冷启动和数据稀疏性问题。2. The present invention utilizes the LDA model to mine relevant information in user video comments, generates a user type latent feature matrix and a video type latent feature matrix, combines the user type latent feature matrix with the video type latent feature matrix to obtain a hidden information matrix, and combines the hidden information matrix with the original user video rating matrix to generate a new user-video rating matrix, thereby alleviating the cold start and data sparsity problems.

附图说明BRIEF DESCRIPTION OF THE DRAWINGS

图1是本发明的方法流程图；Fig. 1 is a flow chart of the method of the present invention;

图2是本发明的多层感知机结构图；FIG2 is a diagram of the structure of a multi-layer perceptron of the present invention;

图3是本发明的深度神经网络结构图。FIG3 is a structural diagram of a deep neural network of the present invention.

具体实施方式DETAILED DESCRIPTION

下面结合说明书附图对本发明的具体实施方式作进一步详细的说明。The specific implementation modes of the present invention will be further described in detail below in conjunction with the accompanying drawings.

本发明是在传统的矩阵分解模型的基础上，重构用户-视频评分矩阵，引入融入用户活跃度的视频关联正则项和可靠的最近邻正则项用来限制潜在特征矩阵的学习。利用深度神经网络的非线性结构，缓解了矩阵分解过程中线性点积的限制，将矩阵分解出的潜在特征向量作为深度神经网络的的输入，经过多层感知机处理得到MLP模型的结果，并在单层感知机层即合并层将MLP模型结果与双正则化矩阵分解模型的结果相结合，通过归一化交叉熵法优化模型，从而提高评分预测的精度。The present invention reconstructs the user-video rating matrix on the basis of the traditional matrix decomposition model, introduces the video association regularization term integrating the user activity and the reliable nearest neighbor regularization term to limit the learning of the potential feature matrix. The nonlinear structure of the deep neural network is utilized to alleviate the limitation of the linear dot product in the matrix decomposition process, and 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 multi-layer perceptron processing, and the MLP model result is combined with the result of the double regularization matrix decomposition model in the single-layer perceptron layer, i.e., the merging layer, and the model is optimized by the normalized cross entropy method, thereby improving the accuracy of the rating prediction.

如图1所示。As shown in Figure 1.

本发明设计了一种基于深度神经网络与双正则化的视频评分预测方法，用于提高对推荐视频评分预测的精度，包括如下步骤：The present invention designs a video rating prediction method based on deep neural network and double regularization to improve the accuracy of recommended video rating prediction, comprising the following steps:

步骤S1：对视频评论进行处理，挖掘出隐藏信息，并将隐藏信息矩阵与原始用户-视频评分矩阵相结合生成新的用户-视频评分矩阵，并进入步骤S2；Step S1: Process the video comments, mine hidden information, and combine the hidden information matrix with the original user-video rating matrix to generate a new user-video rating matrix, and then proceed to step S2;

步骤S2：对用户-视频评分矩阵分解时加入双正则项用来约束潜在特征矩阵的学习，并进入步骤S3；Step S2: When decomposing the user-video rating matrix, a double regularization term is added to constrain the learning of the latent feature matrix, and then proceed to step S3;

步骤S3：将矩阵分解出的潜在特征向量作为多层感知机的输入，经过多层感知机处理得到多层感知机模型预测的结果，并进入步骤S4；Step S3: The potential feature vector decomposed from the matrix is used as the input of the multi-layer perceptron, and the result predicted by the multi-layer perceptron model is obtained after being processed by the multi-layer perceptron, and then the process goes to step S4;

步骤S4：在合并层将多层感知机模型预测的结果与矩阵分解的结果相结合，使用归一化交叉熵法来优化模型，得到最终预测得分。Step S4: In the merging layer, the predicted results of the multilayer perceptron model are combined with the results of matrix decomposition, and the normalized cross entropy method is used to optimize the model to obtain the final prediction score.

具体步骤如下：The specific steps are as follows:

步骤S1包括：首先利用LDA模型挖掘用户视频评论中相关类型的隐藏信息，生成用户类型潜在特征矩阵LU和视频类型潜在特征矩阵LV，并将用户类型潜在特征矩阵与视频类型潜在特征矩阵相结合得到隐藏信息矩阵L，其计算公式为

Figure SMS_43
，并将隐藏信息矩阵L与原始用户-视频评分矩阵R相结合重构用户-视频评分矩阵R，其计算公式为：
Figure SMS_44
。Step S1 includes: firstly, using the LDA model to mine the hidden information of relevant types in user video comments, generating 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, whose calculation formula is:
Figure SMS_43
, and combine the hidden information matrix L with the original user-video rating matrix R to reconstruct the user-video rating matrix R, whose calculation formula is:
Figure SMS_44
.

步骤S2包括：将用户-视频评分矩阵进行矩阵分解，将高维用户-视频评分矩阵分解为低维用户特征矩阵和视频特征矩阵，其公式为：Step S2 includes: performing matrix decomposition on the user-video rating matrix, decomposing the high-dimensional user-video rating matrix into a low-dimensional user feature matrix and a video feature matrix, and the formula is:

Figure SMS_45
Figure SMS_45

其中U表示用户特征矩阵，Ui表示用户i的潜在特征向量，V表示视频特征矩阵，Vj是视频j的潜在特征向量。低维矩阵分解方法通过d秩因子的乘积近似计算评分矩阵R。用户i对视频j的预测评分表示为

Figure SMS_46
，将预测评分与原始评分之间的误差平方作为损失函数，最小化损失函数来逼近评分矩阵R。损失函数为：Where U represents the user feature matrix, Ui represents the potential feature vector of user i, V represents the video feature matrix, and Vj is the potential feature vector of video j. The low-dimensional matrix decomposition method approximates the rating matrix R by multiplying the d-rank factors. The predicted rating of user i for video j is expressed as
Figure SMS_46
, the square error between the predicted score and the original score is used as the loss function, and the loss function is minimized to approximate the score matrix R. The loss function is:

Figure SMS_47
，
Figure SMS_47
,

在上述公式中，

Figure SMS_48
是指示函数，表示如果用户i对项目j进行评分则等于1，否则等于0。
Figure SMS_49
和
Figure SMS_50
为两个正则项，防止过拟合。由于每个用户对视频的评分都会对视频相似度做出一定的贡献，但是每个用户贡献却不完全相同，从用户的活跃度考虑可以分为活跃用户和不活跃用户，活跃用户指的是对视频有大量的评分记录的用户，而不活跃用户是指只对少数视频进行评分记录的用户，所以在计算视频相似度时应将活跃用户和不活跃用户的贡献区分开，用户的活跃度可以定义为：In the above formula,
Figure SMS_48
is an indicator function, which equals 1 if user i rates item j, and 0 otherwise.
Figure SMS_49
and
Figure SMS_50
are two regularization terms to prevent overfitting. Since each user's rating of a video will make a certain contribution to the video similarity, but each user's contribution is not exactly the same, users can be divided into active users and inactive users based on their activity. Active users refer to users who have a large number of rating records for videos, while inactive users refer to users who have only rated a few videos. Therefore, when calculating video similarity, the contributions of active users and inactive users should be distinguished. The activity of users can be defined as:

Figure SMS_51
Figure SMS_51

在上述公式中，

Figure SMS_52
表示用户u的评分总量。因此用户的活跃度系数结合修正的余弦相似度得到的视频相似度计算方法为：In the above formula,
Figure SMS_52
Represents the total number of ratings of user u. Therefore, the video similarity calculation method obtained by combining the user's activity coefficient with the modified cosine similarity is:

Figure SMS_53
Figure SMS_53

在上述公式中，

Figure SMS_54
表示用户u对视频i的评分，
Figure SMS_55
表示用户u对视频j的评分，
Figure SMS_56
表示用户u的评分评分。在矩阵分解时引入融入用户活跃度的视频关联正则项用来约束项目潜在特征矩阵的学习，此时融入用户活跃度的视频关联正则化约束函数公式为：In the above formula,
Figure SMS_54
represents the rating of video i by user u,
Figure SMS_55
represents the rating of video j by user u,
Figure SMS_56
Represents the rating of user u. When decomposing the matrix, the video association regularization term that incorporates user activity is introduced to constrain the learning of the project's latent feature matrix. At this time, the video association regularization constraint function formula that incorporates user activity is:

Figure SMS_57
Figure SMS_57

其中，V表示视频特征矩阵，Vj是视频j的潜在特征向量，Vi是视频i的潜在特征向量。相同兴趣爱好的用户会互相影响，用户相似度可以使用加权的皮尔逊相关系数进行计算：Where V represents the video feature matrix, Vj is the potential feature vector of video j, and Vi is the potential feature vector of video i. Users with the same interests and hobbies will influence each other, and user similarity can be calculated using the weighted Pearson correlation coefficient:

Figure SMS_58
Figure SMS_58

在上述公式中，

Figure SMS_59
和
Figure SMS_60
分别代表用户u和v的平均评分，
Figure SMS_61
为权重即能够影响用户相似度计算的项目的Jaccard相关系数，其计算公式如下：In the above formula,
Figure SMS_59
and
Figure SMS_60
Represent the average ratings of users u and v respectively,
Figure SMS_61
The Jaccard correlation coefficient of the item that can affect the calculation of user similarity is the weight, and its calculation formula is as follows:

Figure SMS_62
Figure SMS_62

其中，

Figure SMS_63
表示用户u评论过的视频集合，
Figure SMS_64
表示用户v评论过的视频集合。in,
Figure SMS_63
represents the set of videos commented by user u,
Figure SMS_64
Represents the set of videos commented by user v.

用户对项目的评分取决于邻近用户的影响，也可能受到邻近用户的邻近用户的影响，但是一定距离后的邻近用户不会在对用户有所影响即变得不在可靠，因此引进可靠值，可靠值大于一定值的邻近用户会对用户项目评分有影响，可靠值计算方式为：The user's rating of an item depends on the influence of neighboring users, and may also be affected by the neighboring users of the neighboring users. However, neighboring users beyond a certain distance will no longer have an impact on the user and become unreliable. Therefore, a reliability value is introduced. Neighboring users with a reliability value greater than a certain value will have an impact on the user's item rating. The reliability value is calculated as follows:

Figure SMS_65
，
Figure SMS_65
,

在上述公式中，

Figure SMS_67
表示用u对视频i的评分，
Figure SMS_69
表示用户v对视频i的评分，
Figure SMS_72
表示用户u评论过的视频集合，
Figure SMS_68
表示用户v评论过的视频集合，
Figure SMS_70
表示评分的最大值，
Figure SMS_71
表示信任距离即用户u和用户v之间存在的用户人数，
Figure SMS_73
表示两个用户之间允许的最大距离，
Figure SMS_66
是为修正参数，是一个大于0小于1的数，。可靠的最近邻用户为In the above formula,
Figure SMS_67
represents the rating of video i by u,
Figure SMS_69
represents the rating of user v on video i,
Figure SMS_72
represents the set of videos commented by user u,
Figure SMS_68
represents the set of videos commented by user v,
Figure SMS_70
represents the maximum value of the score,
Figure SMS_71
represents the trust distance, that is, the number of users between user u and user v.
Figure SMS_73
Indicates the maximum distance allowed between two users.
Figure SMS_66
is a correction parameter, which is a number greater than 0 and less than 1. The reliable nearest neighbor user is

Figure SMS_74
，
Figure SMS_74
,

在矩阵分解时引入可靠的最近邻正则项来约束用户潜在特征矩阵的学习，可靠的最近邻正则项约束函数为：During matrix decomposition, a reliable nearest neighbor regularization term is introduced to constrain the learning of the user's potential feature matrix. The constraint function of the reliable nearest neighbor regularization term is:

Figure SMS_75
Figure SMS_75

其中，

Figure SMS_76
为用户u的潜在向量，
Figure SMS_77
为用户v的潜在特征向量。in,
Figure SMS_76
is the potential vector of user u,
Figure SMS_77
is the potential feature vector of user v.

加入融有用户活跃度的视频关联正则化项和可靠的最近邻正则项，最后的优化损失函数为：By adding the video association regularization term that incorporates user activity and the reliable nearest neighbor regularization term, the final optimization loss function is:

Figure SMS_78
Figure SMS_78

采用随机梯度下降法来寻找最优解，找出最优潜在特征矩阵。The stochastic gradient descent method is used to find the optimal solution and find the optimal latent feature matrix.

如图2所示。As shown in Figure 2.

步骤S3包括：将用户潜在特征向量Uu和视频潜在特征向量Vi作为多层感知机的输入，其中多层感知机如图2所示，包括输入层Lin、若干个允许神经结构非线性的隐藏层和输出层Lout，输入层Lin的输出向量为:Step S3 includes: taking the user potential feature vector U u and the video potential feature vector V i as the input of a multilayer perceptron, wherein the multilayer perceptron is shown in FIG2 , including an input layer L in , a plurality of hidden layers that allow the nonlinearity of the neural structure, and an output layer L out , and the output vector of the input layer L in is:

Figure SMS_79
Figure SMS_79

经过第一个隐藏层处理后输出向量：The output vector after processing by the first hidden layer is:

Figure SMS_80
Figure SMS_80

其中

Figure SMS_81
是包含在输入层和第一层隐藏层L1之间的矩阵中的权重集合，
Figure SMS_82
是L1层的偏差，
Figure SMS_83
是激活函数，激活函数为in
Figure SMS_81
is the set of weights contained in the matrix between the input layer and the first hidden layer L1,
Figure SMS_82
is the bias of the L1 layer,
Figure SMS_83
is the activation function, the activation function is

Figure SMS_84
Figure SMS_84

所以隐藏层Lk的输出向量为：So the output vector of the hidden layer L k is:

Figure SMS_85
Figure SMS_85

其中

Figure SMS_86
为神经元的激活函数，
Figure SMS_87
为权重矩阵，
Figure SMS_88
为偏差。多层感知机输出层Lout输出向量为:in
Figure SMS_86
is the neuron activation function,
Figure SMS_87
is the weight matrix,
Figure SMS_88
is the deviation. The output vector of the multi-layer perceptron output layer L out is:

Figure SMS_89
Figure SMS_89

如图3所示。As shown in Figure 3.

步骤S4包括：在该深度神经网络结构中单层感知机为合并层，其中该深度神经网络结构如图3所示，在合并层将多层感知机模型预测结果与双正则化矩阵分解模型的结果相结合，其计算公式为：Step S4 includes: in the deep neural network structure, the single-layer perceptron is a merging layer, wherein the deep neural network structure is shown in FIG3 , and the prediction result of the multi-layer perceptron model is combined with the result of the double regularization matrix decomposition model in the merging layer, and the calculation formula is:

Figure SMS_90
Figure SMS_90

其中，

Figure SMS_91
为激活函数，
Figure SMS_92
为输出层与合并层之间的矩阵权重集，
Figure SMS_93
为输出层的结果，
Figure SMS_94
为用户潜在向量，
Figure SMS_95
为视频潜在向量，
Figure SMS_96
为合并层的偏差项in,
Figure SMS_91
is the activation function,
Figure SMS_92
is the matrix weight set between the output layer and the merging layer,
Figure SMS_93
is the result of the output layer,
Figure SMS_94
is the user latent vector,
Figure SMS_95
is the video latent vector,
Figure SMS_96
is the bias term of the merging layer

采用归一化交叉熵法通过以下代价函数不断优化所提的出的模型，其代价函数为:The proposed model is continuously optimized by the normalized cross entropy method through the following cost function, and the cost function is:

Figure SMS_97
Figure SMS_97

在上述公式中，

Figure SMS_98
为合并层神经元数量，
Figure SMS_99
为预测出的分数，
Figure SMS_100
为训练实例真实分数，
Figure SMS_101
表示评分的最大值。代价函数使用梯度下降方法不断优化模型，得到最终的预测评分。In the above formula,
Figure SMS_98
is the number of neurons in the merging layer,
Figure SMS_99
is the predicted score,
Figure SMS_100
is the true score of the training instance,
Figure SMS_101
Represents the maximum value of the score. The cost function uses the gradient descent method to continuously optimize the model and obtain the final predicted score.

上面结合附图对本发明的实施方式作了详细说明，但是本发明并不限于上述实施方式，在本领域普通技术人员所具备的知识范围内，还可以在不脱离本发明宗旨的前提下做出各种变化。The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims (2)

1.一种基于深度神经网络与双正则化的视频评分预测方法，其特征在于，所述方法包括如下步骤：1. A video rating prediction method based on deep neural network and double regularization, characterized in that the method comprises the following steps: 步骤S1：对视频评论进行处理，挖掘出隐藏信息，并将隐藏信息矩阵与原始用户-视频评分矩阵相结合生成新的用户-视频评分矩阵，并进入步骤S2，包括：Step S1: Process the video comments to mine hidden information, and combine the hidden information matrix with the original user-video rating matrix to generate a new user-video rating matrix, and then proceed to step S2, including: 首先利用LDA模型挖掘用户视频评论中相关类型的隐藏信息，生成用户类型潜在特征矩阵LU和视频类型潜在特征矩阵LV，并将用户类型潜在特征矩阵与视频类型潜在特征矩阵相结合得到隐藏信息矩阵L，其计算公式为：First, the LDA model is used to mine the hidden information of relevant types in user video comments, 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 hidden information matrix L, whose calculation formula is:
Figure QLYQS_1
Figure QLYQS_1
 并将隐藏信息矩阵L与原始用户-视频评分矩阵R相结合生成新的用户-视频评分矩阵
Figure QLYQS_2
其计算公式为：The hidden information matrix L is combined with the original user-video rating matrix R to generate a new user-video rating matrix
Figure QLYQS_2
The calculation formula is:
Figure QLYQS_3
Figure QLYQS_3
 步骤S2：对用户-视频评分矩阵分解时加入双正则项约束潜在特征矩阵的学习，每个用户对视频的评分都会对视频相似度做出一定的贡献，用户贡献却不完全相同，从用户的活跃度考虑能够分为活跃用户和不活跃用户，活跃用户指的是对视频有大量的评分记录的用户，而不活跃用户是指只对少数视频进行评分记录的用户，所以在计算视频相似度时应将活跃用户和不活跃用户的贡献区分开，用户的活跃度定义为：Step S2: When decomposing the user-video rating matrix, double regularization terms are added to constrain the learning of the latent feature matrix. Each user's rating of a video will make a certain contribution to the video similarity, but the user contributions are not exactly the same. From the perspective of user activity, they can be divided into active users and inactive users. Active users refer to users who have a large number of rating records for videos, while inactive users refer to users who have only rated a few videos. Therefore, when calculating video similarity, the contributions of active users and inactive users should be distinguished. The user's activity is defined as:
Figure QLYQS_4
Figure QLYQS_4
 在公式1中，|M(u)|表示用户u的评分总量，因此用户的活跃度系数结合修正的余弦相似度得到的视频相似度计算方法为：In formula 1, |M(u)| represents the total number of ratings of user u, so the video similarity calculation method obtained by combining the user's activity coefficient with the modified cosine similarity is:
Figure QLYQS_5
Figure QLYQS_5
 在公式2中，Rui表示用户u对视频i的评分，Ruj表示用户u对视频j的评分，
Figure QLYQS_6
表示用户u的平均评分，U表示同时对视频i和j有过评分的用户集合；在矩阵分解时引入融入用户活跃度的视频关联正则项约束项目潜在特征矩阵的学习，此时融入用户活跃度的视频关联正则化约束函数公式为：In Formula 2, Rui represents the rating of video i given by user u, and Ruj represents the rating of video j given by user u.
Figure QLYQS_6
represents the average rating of user u, and U represents the set of users who have rated videos i and j at the same time. When decomposing the matrix, the video association regularization term that incorporates user activity is introduced to constrain the learning of the project's latent feature matrix. At this time, the formula of the video association regularization constraint function that incorporates user activity is:
Figure QLYQS_7
Figure QLYQS_7
 在公式3中，V表示视频特征矩阵，Vj是视频j的潜在特征向量，Vi是视频i的潜在特征向量，并进入步骤S3，包括：In Formula 3, V represents the video feature matrix, Vj is the potential feature vector of video j, Vi is the potential feature vector of video i, and the process proceeds to step S3, including: 相同兴趣爱好的用户会互相影响，用户相似度使用加权的皮尔逊相关系数进行计算：Users with the same interests and hobbies will influence each other, and user similarity is calculated using the weighted Pearson correlation coefficient:
Figure QLYQS_8
Figure QLYQS_8
 在公式6中，
Figure QLYQS_9
和
Figure QLYQS_10
分别代表用户u和v的平均评分，rui表示用户u对视频i的评分，rvi表示用户v对视频i的评分，Iu表示用户u评论过的视频集合，Iv表示用户v评论过的视频集合，JCCuv为权重即能够影响用户相似度计算的项目的Jaccard相关系数，其计算公式如下：In Formula 6,
Figure QLYQS_9
and
Figure QLYQS_10
Represent the average ratings of users u and v respectively, r ui represents the rating of video i by user u, r vi represents the rating of video i by user v, I u represents the set of videos commented on by user u, I v represents the set of videos commented on by user v, JCC uv is the weight, that is, the Jaccard correlation coefficient of the items that can affect the calculation of user similarity, and its calculation formula is as follows:
Figure QLYQS_11
Figure QLYQS_11
 其中，Iu表示用户u评论过的视频集合，Iv表示用户v评论过的视频集合；Where Iu represents the set of videos commented on by user u, and Iv represents the set of videos commented on by user v; 用户对项目的评分取决于邻近用户的影响，也会受到邻近用户的邻近用户的影响，但是一定距离后的邻近用户不会在对用户有所影响即变得不在可靠，因此引进可靠值，可靠值大于一定值的邻近用户会对用户项目评分有影响，可靠值计算方式为：The user's rating of the project depends on the influence of neighboring users, and is also affected by the neighboring users of the neighboring users. However, the neighboring users after a certain distance will no longer have an impact on the user, that is, they become unreliable. Therefore, a reliability value is introduced. Neighboring users with a reliability value greater than a certain value will have an impact on the user's project rating. The reliability value is calculated as follows:
Figure QLYQS_12
Figure QLYQS_12
 在公式8中，Rui表示用u对视频i的评分，Rvi表示用户v对视频i的评分，Iu表示用户u评论过的视频集合，Iv表示用户v评论过的视频集合，Rmax表示评分的最大值，duv表示信任距离即用户u和用户v之间存在的用户人数，dmax表示两个用户之间允许的最大距离，ε是为修正参数，是一个大于0小于1的数，可靠的最近邻用户为In Formula 8, Rui represents the score of video i given by user u, Rvi represents the score of video i given by user v, Iu represents the set of videos commented on by user u, Iv represents the set of videos commented on by user v, Rmax represents the maximum score, duv represents the trust distance, i.e., the number of users between user u and user v, dmax represents the maximum distance allowed between two users, ε is a correction parameter, which is a number greater than 0 and less than 1, and the reliable nearest neighbor user is Near＝{v∈U|simuv≥γ且Cuv≥δ} 公式9Near＝{v∈U|sim uv ≥γ and C uv ≥δ} Formula 9 在矩阵分解时引入可靠的最近邻正则项来约束用户潜在特征矩阵的学习，可靠的最近邻正则项约束函数为：During matrix decomposition, a reliable nearest neighbor regularization term is introduced to constrain the learning of the user's potential feature matrix. The constraint function of the reliable nearest neighbor regularization term is:
Figure QLYQS_13
Figure QLYQS_13
 其中，Uu为用户u的潜在向量，Uv为用户v的潜在特征向量；Among them, U u is the potential vector of user u, and U v is the potential feature vector of user v; 步骤S3：将矩阵分解出的潜在特征向量作为多层感知机的输入，经过多层感知机处理得到多层感知机模型预测的结果，并进入步骤S4；Step S3: The potential feature vector decomposed from the matrix is used as the input of the multi-layer perceptron, and the result predicted by the multi-layer perceptron model is obtained after being processed by the multi-layer perceptron, and then the process goes to step S4; 步骤S4：在合并层将多层感知机模型预测的结果与矩阵分解的结果相结合，使用归一化交叉熵法来优化模型，得到最终预测得分，包括：Step S4: Combine the prediction results of the multi-layer perceptron model with the results of matrix decomposition at the merging layer, use the normalized cross entropy method to optimize the model, and obtain the final prediction score, including: 在该深度神经网络结构中单层感知机为合并层，在合并层将多层感知机模型的预测结果与双正则化矩阵分解模型的结果相结合，其计算公式为：In this deep neural network structure, the single-layer perceptron is a merging layer, in which the prediction results of the multi-layer perceptron model are combined with the results of the double regularization matrix decomposition model. The calculation formula is:
Figure QLYQS_14
Figure QLYQS_14
 在公式11中ρm为激活函数，wm为输出层与合并层之间的矩阵权重集，yout为输出层的结果，Uu为用户潜在向量，Vi为视频潜在向量，am为合并层的偏差项，使用归一化交叉熵法来优化模型，最终得到预测评分。In Formula 11, ρ m is the activation function, w m is the matrix weight set between the output layer and the merging layer, y out is the result of the output layer, U u is the user latent vector, V i is the video latent vector, and a m is the bias term of the merging layer. The normalized cross entropy method is used to optimize the model and finally obtain the predicted score. 2.根据权利要求1所述的一种基于深度神经网络与双正则化的视频评分预测方法，其特征在于：所述步骤S3中，将用户潜在特征向量与视频潜在特征向量作为多层感知机的输入，其中该深度神经网络由多层感知机、单层感知机组成，其中多层感知机包括输入层、若干个允许神经结构非线性的隐藏层和输出层，利用隐藏层的非线性特征，通过多层感知机处理得到多层感知机模型的结果。2. According to claim 1, a video rating prediction method based on deep neural network and double regularization is characterized in that: 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 a multilayer perceptron and a single-layer perceptron, wherein the multilayer perceptron includes an input layer, several hidden layers that allow the nonlinearity of the neural structure, and an output layer, and the nonlinear characteristics of the hidden layer are utilized to obtain the result of the multilayer perceptron model through multilayer perceptron processing.
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 CN115905617A (en) 2023-04-04
CN115905617B true 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)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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

Family Cites Families (8)

* 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
CN111581333B (en) * 2020-05-07 2023-05-26 重庆大学 Text-CNN-based video and audio playlist push method and video and audio playlist push system
CN114117191A (en) * 2020-08-26 2022-03-01 付昳漫 Context-aware recommendation method based on deep neural network
CN112328900A (en) * 2020-11-27 2021-02-05 北京工业大学 A deep learning recommendation method integrating rating matrix and review text
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
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
CN113807422B (en) * 2021-09-07 2024-05-31 南京邮电大学 Weighted graph convolutional neural network scoring prediction model integrating multi-feature information
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

Also Published As

Publication number Publication date
CN115905617A (en) 2023-04-04
WO2024179089A1 (en) 2024-09-06

Similar Documents

Publication Publication Date Title
CN109299396B (en) Convolutional neural network collaborative filtering recommendation method and system fusing attention model
Chen et al. Evolutionary heterogeneous clustering for rating prediction based on user collaborative filtering
Ni et al. An effective recommendation model based on deep representation learning
CN112650929B (en) A Graph Neural Network Recommendation Method Incorporating Review Information
Liu et al. An improved Adam optimization algorithm combining adaptive coefficients and composite gradients based on randomized block coordinate descent
CN110309195B (en) FWDL (full Width Domain analysis) model based content recommendation method
CN110362738A (en) A kind of personalized recommendation method of combination trust and influence power based on deep learning
CN111737578A (en) Recommendation method and system
Chen et al. A survey on heterogeneous one-class collaborative filtering
CN111611375B (en) Text emotion classification method based on deep learning and turning relation
CN110909529A (en) User emotion analysis and prejudgment system of company image promotion system
CN117056609A (en) Session recommendation method based on multi-layer aggregation enhanced contrast learning
Li Cross‐Border E‐Commerce Intelligent Information Recommendation System Based on Deep Learning
CN116071128A (en) A multi-task recommendation method based on multi-behavior feature extraction and self-supervised learning
CN114330291A (en) Text recommendation system based on dual attention mechanism
CN115905617B (en) Video scoring prediction method based on deep neural network and double regularization
CN117390267A (en) Knowledge graph-based personalized multitask enhanced recommendation model
Xu et al. [Retracted] Research on the Construction of Crossborder e‐Commerce Logistics Service System Based on Machine Learning Algorithms
CN115858945A (en) Article recommendation method and system, electronic device and storage medium
CN119003885A (en) Session recommendation method based on self-supervision double-graph and intention association
Deng et al. Group event recommendation based on a heterogeneous attribute graph considering long-and short-term preferences
CN112559905B (en) Conversation recommendation method based on dual-mode attention mechanism and social similarity
CN111368168A (en) Big data-based electricity price obtaining and predicting method, system and computer-readable storage medium
CN117132341A (en) An action sequence recommendation method based on recurrent neural network
Wen et al. Research and Design of Credit Risk Assessment System Based on Big Data and Machine Learning

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