WO2020135193A1 - 一种基于深度神经网络的视频推荐方法、系统及存储介质 - Google Patents

一种基于深度神经网络的视频推荐方法、系统及存储介质 Download PDF

Info

Publication number
WO2020135193A1
WO2020135193A1 PCT/CN2019/126442 CN2019126442W WO2020135193A1 WO 2020135193 A1 WO2020135193 A1 WO 2020135193A1 CN 2019126442 W CN2019126442 W CN 2019126442W WO 2020135193 A1 WO2020135193 A1 WO 2020135193A1
Authority
WO
WIPO (PCT)
Prior art keywords
video
sample data
user
data set
deep neural
Prior art date
Application number
PCT/CN2019/126442
Other languages
English (en)
French (fr)
Inventor
徐永泽
赖长明
Original Assignee
深圳Tcl新技术有限公司
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 深圳Tcl新技术有限公司 filed Critical 深圳Tcl新技术有限公司
Publication of WO2020135193A1 publication Critical patent/WO2020135193A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—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
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4662—Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms
    • H04N21/4666—Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms using neural networks, e.g. processing the feedback provided by the user
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4667—Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies
    • 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
    • H04N21/482—End-user interface for program selection
    • 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
    • H04N21/482—End-user interface for program selection
    • H04N21/4826—End-user interface for program selection using recommendation lists, e.g. of programs or channels sorted out according to their score

Definitions

  • the present disclosure relates to the technical field of display control, and in particular, to a video recommendation method and storage medium based on deep neural networks.
  • recommendation systems can be roughly divided into two categories: contextual recommendation based on browsing information and personalized recommendation based on user historical information.
  • Context-based recommendations consider factors such as location information, date and time.
  • Personalized recommendation usually uses collaborative filtering to recommend products to users.
  • the prediction of user interest is based on the analysis of the tastes and preferences of other users in the system, and implicitly infers the "similarity" between the two. The underlying assumption is that when two people have similar tastes, they will have a higher likelihood of maintaining the same attitude towards the product.
  • the purpose of the present disclosure is to provide a video recommendation method and storage medium based on deep neural networks, to overcome the lack of a user rating mechanism in the prior art when viewing network videos, which cannot be scored according to the user
  • the information is the defect of personalized video recommendation by users.
  • the first embodiment disclosed in the present disclosure is a video recommendation method based on a deep neural network, wherein the personalized recommendation applied to network video includes the following steps:
  • the steps of generating a positive sample data set and a negative sample data set based on the user's viewing video data and the video data of the video library include:
  • a positive sample data set is generated by using all data in the user's viewing video data as positive sample data, and multiple data is extracted from the video library as negative sample data to generate a negative sample data set.
  • the steps of generating a positive sample data set and a negative sample data set based on the user's viewing video data and the video data of the video library include:
  • the steps of generating a positive sample data set and a negative sample data set based on the user's viewing video data and the video data of the video library include:
  • the structure of the self-encoder model is: a self-encoder network structure composed of a fully connected coding layer, a feature layer, and a decoding layer; the coding layer and the decoding layer are symmetrical 2-layer structures, and the The output layer of the self-encoder network structure is transformed into a classifier.
  • the activation function of the output layer of the self-encoder model is a Softmax function
  • the activation function of the middle layer is a scaling exponential linear unit function
  • the step of training the self-encoder model based on the deep neural network using the positive sample data set and the negative sample data set includes:
  • the data in the positive sample data set and the data in the negative sample data set are respectively feature vectorized, and the feature vectorized data are input as training samples to a pre-established self-encoder model based on a deep neural network, and the self-encoder The model is trained.
  • the number of videos extracted in the negative sample data set is related to the total number of videos in the video library.
  • the loss function of the self-encoder model is a cross-entropy loss function.
  • the second embodiment provided by the present disclosure is a video recommendation system based on a deep neural network, which includes:
  • the training set acquisition module is used to acquire the user's viewing video data, and generate a positive sample data set and a negative sample data set according to the user's viewing video data and the video data of the video library;
  • a network model training module used for training the self-encoder model based on the deep neural network using the positive sample data set and the negative sample data set to obtain a training network model
  • the prediction recommendation module is used to input each user's video viewing data and video database video data into the training network model to obtain a predicted video recommendation list;
  • the personalized recommendation module is used to recommend videos to each user based on the predicted recommended video list.
  • the structure of the self-encoder model is: a self-encoder network structure composed of a fully connected coding layer, a feature layer, and a decoding layer; the coding layer and the decoding layer are symmetrical 2-layer structures, and the The output layer of the self-encoder network structure is transformed into a classifier.
  • the activation function of the output layer of the self-encoder model is a Softmax function
  • the activation function of the middle layer is a scaling exponential linear unit function
  • the loss function of the self-encoder model is a cross-entropy loss function.
  • the positive sample data set includes the user watching the complete video
  • the negative sample data set includes the video that the user does not watch.
  • the positive sample data set includes video data whose user viewing completeness exceeds a preset completeness threshold
  • the negative sample data set includes video data whose user viewing completeness is lower than a preset completeness threshold
  • a third embodiment disclosed in the present disclosure is: a storage medium in which a control program for video recommendation based on a deep neural network is stored on the storage medium, and the control program for video recommendation based on a deep neural network is processed by a processor During execution, the steps of the video recommendation method based on the deep neural network described in the item are implemented.
  • the present disclosure provides a deep neural network-based video recommendation method, system, and storage medium, by acquiring the user's viewing video data, and inputting the user's viewing video data as training samples into the pre-established deep neural network-based Self-encoder model to obtain the training network model; feature vectorization of each user's watched video data and video database video data, and input the feature vectorized data into the training network model to obtain the predicted video recommendation list ; Video recommendation for each user based on the predicted recommended video list.
  • the method provided by the present disclosure realizes personalized recommendation under the condition of network television. Since the deep network ensures the accuracy of recommendation, it provides convenience for users to watch videos.
  • FIG. 1 is a flowchart of steps of a video recommendation method based on a deep neural network provided by the present disclosure
  • FIG. 2 is a schematic diagram of a typical 4-layer self-encoder network structure in a deep neural network-based video recommendation method provided by the present disclosure
  • FIG. 3 is a schematic block diagram of the principle of the video recommendation system provided by the present disclosure.
  • a self-encoder is a network structure that implements two conversions, one is to encode (x): R n ⁇ Rd and the other is to decode (x): R d ⁇ R n .
  • the present disclosure provides a video recommendation method based on a deep neural network, so that the video recommendation can be accurately performed for the user based on the user's viewing video data.
  • the first embodiment disclosed in the present disclosure is a video recommendation method based on a deep neural network.
  • the personalized recommendation for network video includes the following steps:
  • Step S1 Obtain the user's viewing video data, and generate a positive sample data set and a negative sample data set according to the user's viewing video data and the video data included in the video library.
  • the viewing video data of multiple users it is first necessary to obtain the viewing video data of multiple users, and use the viewing video data of the users as a positive sample data set.
  • Multiple unwatched video data of multiple users are uniformly extracted from the video library as negative samples to form a negative sample data set to obtain a binary classification data set.
  • the number of videos extracted in the negative sample data set is related to the total number of videos in the video library.
  • the data of the complete video viewing in the recorded video viewing information of the user corresponds to 1
  • the data of the complete video viewing of the user viewing video information corresponds to a constant other than 1.
  • the user can also choose to watch the complete video as the positive sample data set, from the video library Select the video that the user has not watched as the negative sample data set, or other selection methods for positive and negative sample data.
  • the steps of generating a positive sample data set and a negative sample data set based on the user's viewing video data and the video data of the video library include:
  • the video data in which the user's viewing completeness is lower than the preset completeness threshold value in the viewed video data may also be divided into negative sample data sets, for example, watching a video with a completeness of only 5%.
  • video data can also be selected as negative sample data from videos not watched by the user in the video library.
  • Step S2 Use the positive sample data set and the negative sample data set to perform self-encoder model training based on the deep neural network to obtain a training network model.
  • the data in the positive sample data set and the negative sample data set are both feature vectorized, and the positive vector data set and the negative sample data set after feature vectorization are used to train the neural network to obtain the trained network model.
  • the structure of the self-encoder model is: a self-encoder network structure composed of a fully connected coding layer, a feature layer, and a decoding layer; the coding layer and the decoding layer are symmetric 2 Layer structure, and the output layer of the self-encoder network structure is transformed into a classifier, and the activation function of the output layer of the self-encoder model is the Softmax function (Sigmoid function), and the activation function of the middle layer is the scaling index linear unit function.
  • the Softmax function (Sigmoid function) of the classifier property is selected and used in the output layer according to the activation function. Therefore, preferably, the loss function of the self-encoder model is a cross-entropy loss function.
  • Step S3 Input the video data of each user's viewing and the video data of the video library into the training network model to obtain a predicted video recommendation list.
  • Step S4 Perform video recommendation for each user according to the predicted recommended video list.
  • Video recommendation is performed according to the recommendation list predicted for each user in step S3 above.
  • f represents a nonlinear activation function.
  • the activation function of the output layer of the last layer is set to the Softmax function (Sigmoid function), thereby improving the accuracy of the prediction information.
  • the personalized prediction in the prior art is based on collecting user ratings for watching videos, and user ratings are often continuous integers, the most common corresponding loss function should be the equalization error.
  • the average score error it is necessary to calculate only the part where the user has a score record, but not the error of the unscored part.
  • our improvement program does not involve
  • a is a random integer related to the data size.
  • the second embodiment disclosed in the present disclosure is a video recommendation system based on a deep neural network, as shown in FIG. 3, including:
  • the training set acquisition module 310 is used to acquire the user's viewing video data, and generate a positive sample data set and a negative sample data set according to the user's viewing video data and the video data included in the video library; the function is as described in step S1.
  • the network model training module 320 is configured to use the positive sample data set and the negative sample data set to perform self-encoder model training based on a deep neural network to obtain a training network model; its function is as described in step S2.
  • the prediction recommendation module 330 is configured to input each user's video viewing data and video database video data into the training network model to obtain a predicted video recommendation list; its function is as described in step S3.
  • the personality recommendation module 340 is used to recommend videos to each user according to the predicted recommended video list, and the function thereof is as described in step S4.
  • the structure of the self-encoder model is: the structure of the pre-established deep neural network-based self-encoder model is: a 4-layer self-encoder network structure in which the coding layer and the decoding layer are symmetric, and the self-encoding The output layer of its network structure is transformed into a classifier.
  • the loss function of the self-encoder model is a cross-entropy loss function.
  • the method is experimentally verified.
  • the verification step includes:
  • the data watched by the user to the network TV is organized into a feature vector format and generated according to the method described in the table Negative sample data.
  • test data is input into the trained network model to obtain the prediction result, and the prediction result is compared with the correct result of the test data to obtain the test result.
  • a third embodiment disclosed in the present disclosure is: a storage medium in which a control program for video recommendation based on a deep neural network is stored on the storage medium, and the control program for video recommendation based on a deep neural network is processed by a processor During execution, the steps of the video recommendation method based on the deep neural network described in the item are implemented.
  • the storage medium may include a storage program area and a storage data area, wherein the storage program area may store an operating system and application programs required by at least one function; the storage data area may store data used or received by an industrial device management method, and the like.
  • the storage medium may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
  • the above-mentioned computer device can execute the method provided by the embodiments of the present disclosure, and has corresponding function modules for performing the method and obtains corresponding beneficial effects.
  • the above-mentioned computer device can execute the method provided by the embodiments of the present disclosure, and has corresponding function modules for performing the method and obtains corresponding beneficial effects.
  • the present disclosure provides a deep neural network-based video recommendation method, system, and storage medium, by acquiring user viewing video data, and inputting the user's viewing video data as training samples into a pre-established deep neural network-based self-encoder Model, get the training network model; feature vectorize each user's watched video data and video data in the video library, and input the feature vectorized data into the training network model to obtain the predicted video recommendation list; The predicted recommended video list recommends videos for each user.
  • the method provided by the present disclosure realizes personalized recommendation under the condition of network television. Since the deep network ensures the accuracy of recommendation, it provides convenience for users to watch videos.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Human Computer Interaction (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)

Abstract

一种基于深度神经网络的视频推荐方法、系统及存储介质，所述方法包括：获取用户的观看视频数据，根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集（S1）；使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型（S2）；将每个用户观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表（S3）；根据预测出的推荐视频列表对各个用户进行视频推荐（S4）。所述方法实现了网络电视条件下的个性化推荐，且基于深度网络保证了推荐的准确率，为用户观看视频提供了便利。

Description

一种基于深度神经网络的视频推荐方法、系统及存储介质
优先权
所述PCT专利申请要求申请日为2018年12月27日，申请号为201811615808.5的中国专利优先权，本专利申请结合了上述专利的技术方案。
技术领域
本公开涉及显示控制技术领域，尤其涉及的是一种基于深度神经网络的视频推荐方法、及存储介质。
背景技术
目前很多网站都会使用推荐系统给用户推荐产品。推荐系统大致可以划分为两个类别：基于浏览信息的上下文推荐和基于用户历史信息的个性化推荐。基于上下文的推荐考虑了上下文信息诸如位置、日期和时间等因素。个性化推荐通常使用协同滤波的方法向用户推荐产品。在这种方法中，用户兴趣的预测是基于对系统中其他用户的品味和偏好的分析，并隐含地推断出两者之间的“相似性”。潜在的假设是当两个人有相似的品味，他们将有更高的可能性对产品保有相同的态度。
现今已有不少基于深度学习的协同滤波算法方案，其中，包括：能够根据用户对产品的已知评分信息预测用户对未评分过的产品的评分，以此达到推荐目的的方案。这些方案不少利用GPU性能，使得算法可以适用于用户数量和产品数量都很大的情况，且相对快速地训练模型。
现今用户在使用网络电视观看视频节目时，通常只会进行观看的行为，无法或基本不会对视频内容进行评分。这就使得播放平台方无法获得用户对产品的评分信息，也就无法利用评分信息为用户进行个性化视频推荐。但因为我们能够获取用户观看行为的信息，仍然可以通过用户是否观看过某一视频的数据搭建推荐系统。但是考虑到使用网络电视观看视频服务的用户数量庞大，推荐目标视频数量巨大且迭代频率高，现有的基于用户观看行为的推荐系统或是无法提供可观的推荐准确率或是无法解决大数据高频更新的应用需求。
因此，现有技术有待于进一步的改进。
发明内容
鉴于上述现有技术中的不足之处，本公开的目的在于提供一种基于深度神经网络的视频推荐方法、及存储介质，克服现有技术中网络视频观看时缺少用户评分机制导致无法根据用户评分信息为用户进行视频个性化推荐的缺陷。
本公开所公开的第一实施例为一种基于深度神经网络的视频推荐方法，其中，应用于网络视频的个性化推荐，包括以下步骤：
获取用户的观看视频数据，根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集；
使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型；
将每个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表；
根据预测出的推荐视频列表对各个用户进行视频推荐。
可选的，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
将用户的观看视频数据中全部数据作为正样本数据生成正样本数据集，从视频库中抽取多个数据作为负样本数据生成负样本数据集。
可选的，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
将用户的观看视频数据中部分数据作为正样本数据生成正样本数据集，将用户的观看视频数据中部分数据作为负样本数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集。
可选的，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
将用户的观看视频数据中观看视频完整度超出预定阈值的数据作为正样本数据生成正样本数据集和从正样本数据选取后的用户的观看视频数据中选取多个数据作为负样本数据生成负样本数据集。
或者，将用户的观看视频数据中部分数据作为正样本数据生成正样本数据集，将用户的观看视频数据中部分数据作为负样本数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集；
将用户的观看视频数据部分数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集；
或者将用户的观看视频数据中观看视频完整度超出预定阈值的数据作为正样本数据生成正样本数据集和从正样本数据选取后的用户的观看视频数据中选取多个数据作为负样本数据生成负样本数据集。
可选的，所述自编码器模型的结构为：全连接的编码层、特征层和解码层组成的自编码器网络结构；所述编码层和解码层为相对称的2层结构，且所述自编码器网络结构的输出层转变为分类器。
可选的，所述自编码器模型的输出层的激活函数为Softmax函数，中间层的激活函数为缩放指数线性单元函数。
可选的，所述使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练的步骤包括：
分别将正样本数据集中的数据和负样本数据集中的数据进行特征向量化，并将特征向量化的数据作为训练样本输入预先建立的基于深度神经网络的自编码器模型，对所述自编码器模型进行训练。
可选的，所述负样本数据集中被抽取的视频个数与视频库中全部视频数量相关。
可选的，所述自编码器模型的损失函数为交叉熵损失函数。
本公开提供的第二实施例为一种基于深度神经网络的视频推荐系统，其中，包括：
训练集获取模块，用于获取用户的观看视频数据，根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集；
网络模型训练模块，用于使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型；
预测推荐模块，用于将每个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表；
个性推荐模块，用于根据预测出的推荐视频列表对各个用户进行视频推荐。
可选的，所述自编码器模型的结构为：全连接的编码层、特征层和解码层组成的自编码器网络结构；所述编码层和解码层为相对称的2层结构，且所述自编码器网络结构的输出层转变为分类器。
可选的，所述自编码器模型的输出层的激活函数为Softmax函数，中间层的激活函数为缩放指数线性单元函数。
可选的，所述自编码器模型的损失函数为交叉熵损失函数。
可选的，所述正样本数据集包括用户观看完整的视频，负样本数据集包括用户未观看的视频。
可选的，所述正样本数据集包括用户观看完整度超出预设完整度阈值的视频数据，负样本数据集包括用户观看完整度低于预设完整度阈值的视频数据。
本公开公开的第三实施例为：一种存储介质，其中，所述存储介质上存储有基于深度神经网络的视频推荐的控制程序，所述基于深度神经网络的视频推荐的控制程序被处理器执行时实现项所述的基于深度神经网络的视频推荐方法的步骤。
有益效果，本公开提供了一种基于深度神经网络的视频推荐方法、系统及存储介质，通过获取用户的观看视频数据，并将用户的观看视频数据作为训练样本输入预先建立的基于深度神经网络的自编码器模型，得到训练网络模型；将每个用户的观看视频数据和视频库的视频数据进行特征向量化，并将特征向量化的数据输入所述训练网络模型，得到预测出的视频推荐列表；根据预测出的推荐视频列表对各个用户进行视频推荐。本公开所提供的方法实现网络电视条件下的个性化推荐，由于基于深度网络保证了推荐的准确率，为用户观看视频提供了便利。
附图说明
图1是本公开所提供的基于深度神经网络的视频推荐方法的步骤流程图；
图2是本公开所提供的基于深度神经网络的视频推荐方法中典型的4层自编码器网络结构示意图；
图3是本公开所提供的所述视频推荐系统的原理示意框图。
具体实施方式
为使本公开的目的、技术方案及优点更加清楚、明确，以下参照附图并举实施例对本公开进一步详细说明。应当理解，此处所描述的具体实施例仅仅用于解释本公开，并不用于限定本公开。
自编码器是一种实现了两个转换的网络结构，一个转换为编码encode(x):R n→R d另一个为解码decode(x):R d→R n。自编码器的目标是得到维度为d的数据特征使得x和f(x)＝decode(encode(x))之间的误差最小化。基于自编码器模型，本公开提供 了一种基于深度神经网络的视频推荐方法，使得可以基于用户的观看视频数据准确的为用户进行视频推荐。
实施例1
本公开所公开的第一实施例为一种基于深度神经网络的视频推荐方法，如图1所示，应用于网络视频的个性化推荐，包括以下步骤：
步骤S1、获取用户的观看视频数据，根据用户的观看视频数据和视频库中包括的视频数据生成正样本数据集和负样本数据集。
为了给每个用户进行视频个性化推荐，首先需要获取用户的历史观看视频信息，并给从历史观看视频信息中得到其所喜爱的视频类型，其观看视频的情况，并根据历史观看视频信息为其进行视频推荐。因此，本步骤首先需要获取多个用户的观看视频数据，将用户的观看视频数据作为正样本数据集。从视频库中均匀抽取多个用户未观看视频数据作为负样本，组成负样本数据集，得到二分类数据集。具体的，所述负样本数据集中被抽取的视频个数与视频库中全部视频数量相关。
为了便于对用户观看信息进行记录，具体的，记录用户的观看视频信息中视频观看完整的数据对应为1，用户观看视频信息中视频未观看完整的数据对应为非1的常数。
可以想到的是，在进行正样本数据集和负样本数据集的选择上，为了达到不同的推荐效果，在具体实施例中，还可以选择用户观看完整的视频作为正样本数据集，从视频库中选择用户未观看的视频作为负样本数据集，或者其他正负样本数据的选择方式。
所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
将用户的观看视频数据中全部数据作为正样本数据生成正样本数据集，从视频库中抽取多个数据作为负样本数据生成负样本数据集；
或者，将用户的观看视频数据中部分数据作为正样本数据生成正样本数据集，将用户的观看视频数据中部分数据作为负样本数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集；
将用户的观看视频数据部分数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集；
或者将用户的观看视频数据中观看视频完整度超出预定阈值的数据作为正样本数据生成正样本数据集和从正样本数据选取后的用户的观看视频数据中选取多个数据作为负样本数据生成负样本数据集。
即是，可以将全部的用户的观看视频数据作为正样本数据集，也可以选取其中部分观看时间超出预设完整度的视频作为正样本数据集，比如：选取其中观看完整度为50％的视频作为正样本数据，也可以将观看视频数据中用户观看完整度低于预设完整度阈值的视频数据划分为负样本数据集，比如：观看完整度仅仅为5％的视频。当然也可以从视频库中用户未观看的视频中选取视频数据作为负样本数据。
步骤S2、使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型。
将正样本数据集中的数据和负样本数据集中的数据均进行特征向量化，并使用特征向量化后的正样本数据集和负样本数据集进行神经网络的训练，得到训练后的网络模型。
具体的，结合图2所示，所述自编码器模型的结构为：全连接的编码层、特征层和解码层组成的自编码器网络结构；所述编码层和解码层为相对称的2层结构，且所述自编码器网络结构的输出层转变为分类器，且所述自编码器模型的输出层的激活函数为Softmax函数(Sigmoid函数)，中间层的激活函数为缩放指数线性单元函数。相应的，根据激活函数在输出层选择使用分类器性质的Softmax函数(Sigmoid函数)，因此，较佳的，所述自编码器模型的损失函数为交叉熵损失函数。
步骤S3、将各个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表。
将各个用户的观看视频数据和视频库的视频数据按照自编码器网络结构的格式进行特征向量化之后，输入上述步骤中训练得到的网络模型，得到为各个用户预测出的视频推荐列表。
步骤S4、根据预测出的推荐视频列表对各个用户进行视频推荐。
根据上述步骤S3中为每个用户预测出的推荐列表进行视频推荐。
由于在原始模型中，无论是编码还是解码部分使用的都是经典的向前传播的全连接神经网络l＝f(W*x+b),这里f表示非线性的激活函数。但是如果激活函数的值域不能包含数据本身的范围，decode的最后一层还是会选择线性函数连接。非线性函数的类型根据目标问题的情况进行选择。因此本公开所提供的方法中，将最后一层的输出层的激活函数设置为Softmax函数(Sigmoid函数)，从而提高了预测信息的准确性。
进一步的，因为现有技术中进行个性化预测时，都是通过收集用户对观看视频的打分进行预测，而用户的打分常常是连续的整数，最常对应的损失函数应为均分误差，在计算均分误差时，需要只计算用户有打分记录的部分，而不计算未打分部分的误差，预 测的结果存在一定的误差，而本公开所提供的方法，我们的改进方案是在不涉及到打分记录的前提下，从用户未观看视频或观看时间超出预设时间值的视频集中均匀地随机抽取a部视频信息，作为负样本使用，改变其ri＝μ。其中a是一个随机整数与数据尺寸相关。μ是一个不等于1的常值，如μ＝-1。这样我们就得到了有正负样本的二分类数据，从而提高了预测准确性。
实施例2
本公开所公开的第二实施例为一种基于深度神经网络的视频推荐系统，如图3所示，包括：
训练集获取模块310，用于获取用户的观看视频数据，根据用户的观看视频数据和视频库中包括的视频数据生成正样本数据集和负样本数据集；其功能如步骤S1所述。
网络模型训练模块320，用于使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型；其功能如步骤S2所述。
预测推荐模块330，用于将每个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表；其功能如步骤S3所述。
个性推荐模块340，用于根据预测出的推荐视频列表对各个用户进行视频推荐，其功能如步骤S4所述。
具体的，所述自编码器模型的结构为：预先建立的基于深度神经网络的自编码器模型的结构为：编码层和解码层相对称的4层自编码器网络结构，且所述自编码其网络结构的输出层转变为分类器。
所述自编码器模型的损失函数为交叉熵损失函数。
为了验证本公开所述方法及系统可以实现为用户进行较为准确的视频推荐，将本方法进行实验验证。
具体的，该验证步骤包括：
采集部分用户采集的部分用户观看网络电视的数据，整理数据，按照本公开所提供的5层Autoencoder网络输入格式需要，将用户观看网络电视的数据整理成特征向量格式并照表格中阐述方法生成正负样本数据。
使用采集到的正负样本数据训练网络模型，得到训练后的网络模型。
将测试数据输入训练后的网络模型，得到预测结果，并将预测结果与测试数据的正确结果相比较，得到得出测试结果。
在本次测试中，针对每个用户从所有的视频中随机抽取99部视频，加上测试目标 的一部视频组成100部视频的集合。以这个集合为推荐视频集进行推荐视频目录预测，通过本公开所提供的方法进行预测后，得到目标视频在推荐列表的前10位的比例为0.9499。
Figure PCTCN2019126442-appb-000001
实施例3
本公开公开的第三实施例为：一种存储介质，其中，所述存储介质上存储有基于深度神经网络的视频推荐的控制程序，所述基于深度神经网络的视频推荐的控制程序被处 理器执行时实现项所述的基于深度神经网络的视频推荐方法的步骤。
存储介质可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储工业设备管理方法使用或接收的数据等。此外，存储介质可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。
上述计算机设备可执行本公开实施例所提供的方法，具备执行该方法相应的功能模块和得到相应的有益效果。未在本实施例中详尽描述的技术细节，可参见本公开实施例所提供的方法。
本公开提供了一种基于深度神经网络的视频推荐方法、系统及存储介质，通过获取用户的观看视频数据，并将用户的观看视频数据作为训练样本输入预先建立的基于深度神经网络的自编码器模型，得到训练网络模型；将每个用户的观看视频数据和视频库中的视频数据进行特征向量化，并将特征向量化的数据输入所述训练网络模型，得到预测出的视频推荐列表；根据预测出的推荐视频列表对各个用户进行视频推荐。本公开所提供的方法实现网络电视条件下的个性化推荐，由于基于深度网络保证了推荐的准确率，为用户观看视频提供了便利。
可以理解的是，对本领域普通技术人员来说，可以根据本公开的技术方案及其发明构思加以等同替换或改变，而所有这些改变或替换都应属于本公开所附的权利要求的保护范围。

Claims (15)

  1. 一种基于深度神经网络的视频推荐方法，其特征在于，应用于网络视频的个性化推荐，包括以下步骤：
    获取用户的观看视频数据，根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集；
    使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型；
    将每个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表；
    根据预测出的推荐视频列表对各个用户进行视频推荐。
  2. 根据权利要求1所述的基于深度神经网络的视频推荐方法，其特征在于，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
    将用户的观看视频数据中全部数据作为正样本数据生成正样本数据集，从视频库中抽取多个数据作为负样本数据生成负样本数据集。
  3. 根据权利要求1所述的基于深度神经网络的视频推荐方法，其特征在于，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
    将用户的观看视频数据中部分数据作为正样本数据生成正样本数据集，将用户的观看视频数据中部分数据作为负样本数据和从视频库中抽取多个数据作为负样本数据生成负样本数据集。
  4. 根据权利要求1所述的基于深度神经网络的视频推荐方法，其特征在于，所述根据用户的观看视频数据和视频库的视频数据生成正样本数据集和负样本数据集的步骤包括：
    将用户的观看视频数据中观看视频完整度超出预定阈值的数据作为正样本数据生成正样本数据集和从正样本数据选取后的用户的观看视频数据中选取多个数据作为负样本数据生成负样本数据集。
  5. 根据权利要求1所述的基于深度神经网络的视频推荐方法，其特征在于，所述自编码器模型的结构为：全连接的编码层、特征层和解码层组成的自编码器网络结构；所述编码层和解码层为相对称的2层结构，且所述自编码器网络结构的输出层转变为分 类器。
  6. 根据权利要求2所述的基于深度神经网络的视频推荐方法，其特征在于，所述自编码器模型的输出层的激活函数为Softmax函数，中间层的激活函数为缩放指数线性单元函数。
  7. 根据权利要求6所述的基于深度神经网络的视频推荐方法，其特征在于，所述使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练的步骤包括：
    分别将正样本数据集中的数据和负样本数据集中的数据进行特征向量化，并将特征向量化的数据作为训练样本输入预先建立的基于深度神经网络的自编码器模型，对所述自编码器模型进行训练。
  8. 根据权利要求7所述的基于深度神经网络的视频推荐方法，其特征在于，所述自编码器模型的损失函数为交叉熵损失函数。
  9. 一种基于深度神经网络的视频推荐系统，其特征在于，包括：
    训练集获取模块，用于获取用户的观看视频数据，根据用户的观看视频数据和视频库中包括的视频数据生成正样本数据集和负样本数据集；
    网络模型训练模块，用于使用所述正样本数据集和负样本数据集进行基于深度神经网络的自编码器模型训练，得到训练网络模型；
    预测推荐模块，用于将每个用户的观看视频数据和视频库的视频数据输入所述训练网络模型，得到预测出的视频推荐列表；
    个性推荐模块，用于根据预测出的推荐视频列表对各个用户进行视频推荐。
  10. 根据权利要求9基于深度神经网络的视频推荐系统，其特征在于，所述自编码器模型的结构为：全连接的编码层、特征层和解码层组成的自编码器网络结构；所述编码层和解码层为相对称的2层结构，且所述自编码器网络结构的输出层转变为分类器。
  11. 根据权利要求10基于深度神经网络的视频推荐系统，其特征在于，所述自编码器模型的输出层的激活函数为Softmax函数，中间层的激活函数为缩放指数线性单元函数。
  12. 根据权利要求10基于深度神经网络的视频推荐系统，其特征在于，所述自编码器模型的损失函数为交叉熵损失函数。
  13. 根据权利要求9基于深度神经网络的视频推荐系统，其特征在于，所述正样本数据集包括用户观看完整的视频，负样本数据集包括用户未观看的视频。
  14. 根据权利要求9基于深度神经网络的视频推荐系统，其特征在于，所述正样本数据集包括用户观看完整度超出预设完整度阈值的视频数据，负样本数据集包括用户观看完整度低于预设完整度阈值的视频数据。
  15. 一种存储介质，其特征在于，所述存储介质上存储有基于深度神经网络的视频推荐的控制程序，所述基于深度神经网络的视频推荐的控制程序被处理器执行时实现如权利要求1至8中任一项所述的基于深度神经网络的视频推荐方法的步骤。
PCT/CN2019/126442 2018-12-27 2019-12-19 一种基于深度神经网络的视频推荐方法、系统及存储介质 WO2020135193A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811615808.5A CN111382307B (zh) 2018-12-27 2018-12-27 一种基于深度神经网络的视频推荐方法、系统及存储介质
CN201811615808.5 2018-12-27

Publications (1)

Publication Number Publication Date
WO2020135193A1 true WO2020135193A1 (zh) 2020-07-02

Family

ID=71129180

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/126442 WO2020135193A1 (zh) 2018-12-27 2019-12-19 一种基于深度神经网络的视频推荐方法、系统及存储介质

Country Status (2)

Country Link
CN (1) CN111382307B (zh)
WO (1) WO2020135193A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931058A (zh) * 2020-08-19 2020-11-13 中国科学院深圳先进技术研究院 一种基于自适应网络深度的序列推荐方法和系统
CN112507163A (zh) * 2020-12-02 2021-03-16 北京奇艺世纪科技有限公司 时长预测模型训练方法、推荐方法、装置、设备及介质
CN112966148A (zh) * 2021-03-05 2021-06-15 安徽师范大学 基于深度学习和特征融合的视频推荐方法和系统
CN113688281A (zh) * 2021-07-09 2021-11-23 温州大学 一种基于深度学习行为序列的视频推荐方法及系统
CN114371002A (zh) * 2021-12-30 2022-04-19 天津理工大学 基于dae-cnn的行星齿轮箱故障诊断方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112650884B (zh) * 2020-12-21 2022-04-01 四川长虹电器股份有限公司 基于深度学习的视频推送方法
CN113220936B (zh) * 2021-06-04 2023-08-15 黑龙江广播电视台 基于随机矩阵编码和简化卷积网络的视频智能推荐方法、装置及存储介质
CN113556484B (zh) * 2021-07-16 2024-02-06 北京达佳互联信息技术有限公司 视频处理方法、装置、电子设备及计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040153373A1 (en) * 2003-01-31 2004-08-05 Docomo Communications Laboratories Usa, Inc. Method and system for pushing services to mobile devices in smart environments using a context-aware recommender
CN102073864A (zh) * 2010-12-01 2011-05-25 北京邮电大学 四层结构的体育视频中足球项目检测系统及实现
CN105160008A (zh) * 2015-09-21 2015-12-16 合一网络技术(北京)有限公司 一种定位推荐用户的方法及装置
US20160188671A1 (en) * 2014-12-29 2016-06-30 Facebook, Inc. Methods and Systems for Recommending Applications
CN108243357A (zh) * 2018-01-25 2018-07-03 北京搜狐新媒体信息技术有限公司 一种视频推荐方法及装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170169330A1 (en) * 2015-12-15 2017-06-15 Le Holdings (Beijing) Co., Ltd. Method and Electronic Device for Displaying Play Content in Smart Television
CN107515909B (zh) * 2017-08-11 2020-05-19 深圳市云网拜特科技有限公司 一种视频推荐方法及系统
CN108959603B (zh) * 2018-07-13 2022-03-29 北京印刷学院 基于深度神经网络的个性化推荐系统及方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040153373A1 (en) * 2003-01-31 2004-08-05 Docomo Communications Laboratories Usa, Inc. Method and system for pushing services to mobile devices in smart environments using a context-aware recommender
CN102073864A (zh) * 2010-12-01 2011-05-25 北京邮电大学 四层结构的体育视频中足球项目检测系统及实现
US20160188671A1 (en) * 2014-12-29 2016-06-30 Facebook, Inc. Methods and Systems for Recommending Applications
CN105160008A (zh) * 2015-09-21 2015-12-16 合一网络技术(北京)有限公司 一种定位推荐用户的方法及装置
CN108243357A (zh) * 2018-01-25 2018-07-03 北京搜狐新媒体信息技术有限公司 一种视频推荐方法及装置

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931058A (zh) * 2020-08-19 2020-11-13 中国科学院深圳先进技术研究院 一种基于自适应网络深度的序列推荐方法和系统
CN111931058B (zh) * 2020-08-19 2024-01-05 中国科学院深圳先进技术研究院 一种基于自适应网络深度的序列推荐方法和系统
CN112507163A (zh) * 2020-12-02 2021-03-16 北京奇艺世纪科技有限公司 时长预测模型训练方法、推荐方法、装置、设备及介质
CN112507163B (zh) * 2020-12-02 2023-07-21 北京奇艺世纪科技有限公司 时长预测模型训练方法、推荐方法、装置、设备及介质
CN112966148A (zh) * 2021-03-05 2021-06-15 安徽师范大学 基于深度学习和特征融合的视频推荐方法和系统
CN113688281A (zh) * 2021-07-09 2021-11-23 温州大学 一种基于深度学习行为序列的视频推荐方法及系统
CN113688281B (zh) * 2021-07-09 2023-11-17 温州大学 一种基于深度学习行为序列的视频推荐方法及系统
CN114371002A (zh) * 2021-12-30 2022-04-19 天津理工大学 基于dae-cnn的行星齿轮箱故障诊断方法
CN114371002B (zh) * 2021-12-30 2024-01-09 天津理工大学 基于dae-cnn的行星齿轮箱故障诊断方法

Also Published As

Publication number Publication date
CN111382307B (zh) 2023-10-17
CN111382307A (zh) 2020-07-07

Similar Documents

Publication Publication Date Title
WO2020135193A1 (zh) 一种基于深度神经网络的视频推荐方法、系统及存储介质
US10460247B2 (en) Attribute weighting for media content-based recommendation
US10824941B2 (en) End-to-end deep collaborative filtering
CN110704674B (zh) 一种视频播放完整度预测方法及装置
EP3055790B1 (en) System, method, and computer program product for providing contextually-aware video recommendation
CN112464100B (zh) 信息推荐模型训练方法、信息推荐方法、装置及设备
CN112734104B (zh) 一种融合生成对抗网络和自编码器的跨域推荐方法
CN112163149B (zh) 推荐消息的方法及装置
CN114357201B (zh) 基于信息感知的视听推荐方法、系统
US20210200799A1 (en) System and method for improved content discovery
Li et al. TPFN: Applying outer product along time to multimodal sentiment analysis fusion on incomplete data
CN117251622A (zh) 对象推荐的方法、装置、计算机设备和存储介质
US11868857B2 (en) Video augmentation apparatus and a method for its use
Mohamad et al. Collaborative filtering approach for movie recommendations
Eden et al. Investigating the Value of Subtitles for Improved Movie Recommendations
Xu et al. How Does the System Perceive Me?—A Transparent and Tunable Recommender System
Bhalgat FusedLSTM: Fusing frame-level and video-level features for content-based video relevance prediction
CN113468415B (zh) 一种融合电影属性和交互信息的电影推荐系统
CN116523024B (zh) 召回模型的训练方法、装置、设备及存储介质
KR102615165B1 (ko) 콘텐츠 스트리밍 시스템에서 유사 콘텐츠를 제공하기 위한 방법 및 장치
Kim et al. Developing a deep learning-based recommendation model using online reviews for predicting consumer preferences: Evidence from the restaurant industry
CN112487300A (zh) 一种视频推荐方法、装置、电子设备及存储介质
Speciale et al. Recommendation Systems in Libraries: an Application with Heterogeneous Data Sources
Chawade et al. The Book Forum: Recommendation Application System using Collaborative Filtering and Autoencoders
CN116955782A (zh) 跨域推荐模型的训练方法、装置、设备、介质及产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19901438

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19901438

Country of ref document: EP

Kind code of ref document: A1