CN110598016A - Method, device, equipment and medium for recommending multimedia information - Google Patents

Method, device, equipment and medium for recommending multimedia information Download PDF

Info

Publication number
CN110598016A
CN110598016A CN201910857435.0A CN201910857435A CN110598016A CN 110598016 A CN110598016 A CN 110598016A CN 201910857435 A CN201910857435 A CN 201910857435A CN 110598016 A CN110598016 A CN 110598016A
Authority
CN
China
Prior art keywords
user
multimedia information
feedback
recommended
portrait
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910857435.0A
Other languages
Chinese (zh)
Other versions
CN110598016B (en
Inventor
伍楚涵
白肇强
白雪峰
程文文
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.)
Shenzhen Yayue Technology Co ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN201910857435.0A priority Critical patent/CN110598016B/en
Publication of CN110598016A publication Critical patent/CN110598016A/en
Application granted granted Critical
Publication of CN110598016B publication Critical patent/CN110598016B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/48—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
    • G06F16/9535—Search customisation based on user profiles and personalisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Multimedia (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The method for recommending the multimedia information comprises the steps of screening the acquired multimedia information according to preset screening conditions, acquiring multimedia information labels of screened candidate multimedia information, and predicting a feedback result of a user to be recommended on the candidate multimedia information through a feedback prediction model and a user portrait which are arranged corresponding to the multimedia information labels. And recommending the candidate multimedia information to the user to be recommended only when the feedback result represents non-negative feedback, and otherwise, filtering the candidate multimedia information. Therefore, after the multimedia information is preliminarily screened, the recommended multimedia information is screened again according to the estimated feedback result of the user to the multimedia information, the optimization of the multimedia information recommendation is realized, and the negative feedback of the user is reduced.

Description

Method, device, equipment and medium for recommending multimedia information
Technical Field
The present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a device, and a medium for recommending multimedia information.
Background
With the development of internet technology, the amount of multimedia information is increasing. In order to improve user experience, a multimedia information application will generally recommend relevant multimedia information to a user according to the multimedia information currently viewed by the user, and after receiving negative feedback from the user on the recommended multimedia information, shield the source of the multimedia information for the user.
For example, a video application obtains a relevant video according to a video currently watched by a user, recommends the relevant video to the user, and masks a video source for the user when receiving a negative evaluation of the user.
However, in the prior art, when determining to recommend multimedia information, each piece of multimedia information is simply screened only according to specific conditions such as the multimedia information currently viewed by a user, the recommendation accuracy of the multimedia information is low, which may cause negative evaluation of the user, and after the multimedia information is applied to receive negative feedback of the user, the multimedia information recommendation cannot be accurately adjusted according to the received negative feedback, and similar multimedia information that is not interested by the user may still be recommended to the user. Therefore, how to optimize multimedia information recommendation and reduce negative feedback of users is a problem to be solved urgently.
Disclosure of Invention
The embodiment of the application provides a method, a device, equipment and a medium for recommending multimedia information, which are used for optimizing the recommendation of the multimedia information when the multimedia information is recommended to a user, so that the recommendation accuracy of the multimedia information is improved.
In one aspect, a method for recommending multimedia information is provided, including:
screening the acquired multimedia information according to preset screening conditions to obtain candidate multimedia information to be recommended;
acquiring a multimedia information tag of candidate multimedia information to be recommended;
acquiring a feedback estimation model set corresponding to a multimedia information tag and a user portrait of a user to be recommended, wherein the feedback estimation model corresponding to the multimedia information tag is used for predicting the feedback of the user to the multimedia information corresponding to the multimedia information tag according to the user portrait of the user;
according to the feedback estimation model and the user portrait, estimating a feedback result of the user to be recommended to the candidate multimedia information;
when the feedback result represents negative feedback, the candidate multimedia information is not recommended to the user to be recommended;
and when the feedback result represents non-negative feedback, recommending candidate multimedia information to the user to be recommended.
In one aspect, an apparatus for multimedia information recommendation is provided, including:
the screening unit is used for screening the acquired multimedia information according to preset screening conditions to obtain candidate multimedia information to be recommended;
the first obtaining unit is used for obtaining a multimedia information label of candidate multimedia information to be recommended;
the second obtaining unit is used for obtaining a feedback estimation model set corresponding to the multimedia information label and a user portrait of a user to be recommended, and the feedback estimation model corresponding to one multimedia information label is used for predicting the feedback of the user to the multimedia information corresponding to the multimedia information label according to the user portrait of the user;
the pre-estimation unit is used for pre-estimating the feedback result of the user to be recommended to the candidate multimedia information according to the feedback pre-estimation model and the user portrait;
and the recommending unit is used for not recommending the candidate multimedia information to the user to be recommended when the feedback result represents negative feedback, and recommending the candidate multimedia information to the user to be recommended when the feedback result represents non-negative feedback.
In one aspect, a control device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of any of the methods for recommending multimedia information.
In one aspect, a computer-readable storage medium is provided, on which a computer program is stored, which, when being executed by a processor, implements the steps of any of the above-mentioned methods for multimedia information recommendation.
In the method, the device, the equipment and the medium for recommending the multimedia information, the obtained multimedia information is screened according to the preset screening condition, and candidate multimedia information to be recommended is obtained; and acquiring a multimedia information label of the candidate multimedia information to be recommended, and predicting a feedback result of the candidate multimedia information of the user to be recommended through a feedback prediction model and a user portrait which are set corresponding to the multimedia information label. And recommending the candidate multimedia information to the user to be recommended only when the feedback result represents non-negative feedback, and otherwise, filtering the candidate multimedia information. Therefore, each multimedia information is preliminarily screened through the preset screening condition, then the recommended multimedia information is screened again through the estimated feedback result of the user on the screened candidate multimedia information, the multimedia information recommendation is optimized, the multimedia information recommendation accuracy is improved, the multimedia information recommendation which is not interesting to the user is reduced, the negative feedback rate of the user is reduced, and the user experience is improved.
Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
FIG. 1a is a schematic diagram illustrating an architecture of multimedia information recommendation according to an embodiment of the present application;
FIG. 1b is a diagram of a multimedia information quality inspection module according to an embodiment of the present disclosure;
fig. 1c is an application scenario of multimedia information recommendation according to an embodiment of the present application;
FIG. 2 is a flowchart illustrating an implementation of a method for training a feedback prediction model according to an embodiment of the present disclosure;
FIG. 3a is a diagram illustrating an example of a user image according to an embodiment of the present application;
FIG. 3b is a schematic diagram of a feedback prediction model according to an embodiment of the present disclosure;
fig. 4 is a flowchart illustrating an implementation of multimedia information recommendation according to an embodiment of the present application;
FIG. 5 is a schematic structural diagram of an apparatus for recommending multimedia information according to an embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of a control device in an embodiment of the present application.
Detailed Description
In order to make the purpose, technical solution and beneficial effects of the present application more clear and more obvious, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
First, some terms referred to in the embodiments of the present application are explained to facilitate understanding by those skilled in the art.
Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.
The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.
Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.
Multimedia information: a man-machine interactive information exchange and transmission medium combining two or more media. The media includes text, pictures, sound, movies, etc.
User portrait: the user portrait is also called as a user role, and is widely applied to various fields as an effective tool for delineating target users and connecting user appeal and design direction. In the actual operation process, the attributes, behaviors and expectations of the user are usually combined by the words which are most shallow and close to life. As a virtual representation of an actual user, a user role formed by user images is constructed according to multimedia operation data and preferences of the user in a period of time, and the formed user role needs to represent a main audience and a target group of products.
Convolutional Neural Network (CNN): the method is a feedforward neural network, and the artificial neurons of the feedforward neural network can respond to peripheral units in a part of coverage range and have excellent performance on large-scale image processing.
Principal Component Analysis (PCA): the high-dimensional data set can be mapped to a low-dimensional space, information loss is reduced as much as possible, and difference among data is reserved. The PCA rotation data set is aligned with its principal components, retaining the most variables in the first principal component.
Fully connected layers (FC): and the convolutional neural network plays a role of a classifier. If we say that operations such as convolutional layers, pooling layers, and activation function layers map raw data to hidden layer feature space, the fully-connected layer serves to map the learned "distributed feature representation" to the sample label space. In practical use, the fully-connected layer may be implemented by a convolution operation: a fully-connected layer that is fully-connected to the previous layer may be converted to a convolution with a convolution kernel of 1x 1; and the fully-connected layer of which the front layer is the convolution layer can be converted into the global convolution with the convolution kernel h x w, wherein h and w are the height and width of the convolution result of the front layer respectively.
Softmax function: also referred to as normalized exponential function, is a generalization of activation functions. The Softmax function is actually a log-gradient normalization of the finite discrete probability distribution.
Activation Function (Activation Function): the function which runs on the neuron of the artificial neural network is responsible for mapping the input of the neuron to the output end, and has very important function for the artificial neural network model to learn and understand the very complex and nonlinear function. They introduce non-linear characteristics into our network.
Desensitization: and (3) converting the real sensitive data which is easy to divulge the secret into non-real data which is not easy to divulge the secret according to a certain rule.
The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism and an encryption algorithm. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product services layer, and an application services layer.
The block chain underlying platform can comprise processing modules such as user management, basic service, intelligent contract and operation monitoring. The user management module is responsible for identity information management of all blockchain participants, and comprises public and private key generation maintenance (account management), key management, user real identity and blockchain address corresponding relation maintenance (authority management) and the like, and under the authorization condition, the user management module supervises and audits the transaction condition of certain real identities and provides rule configuration (wind control audit) of risk control; the basic service module is deployed on all block chain node equipment and used for verifying the validity of the service request, recording the service request to storage after consensus on the valid request is completed, for a new service request, the basic service firstly performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts service information (consensus management) through a consensus algorithm, transmits the service information to a shared account (network communication) completely and consistently after encryption, and performs recording and storage; the intelligent contract module is responsible for registering and issuing contracts, triggering the contracts and executing the contracts, developers can define contract logics through a certain programming language, issue the contract logics to a block chain (contract registration), call keys or other event triggering and executing according to the logics of contract clauses, complete the contract logics and simultaneously provide the function of upgrading and canceling the contracts; the operation monitoring module is mainly responsible for deployment, configuration modification, contract setting, cloud adaptation in the product release process and visual output of real-time states in product operation, such as: alarm, monitoring network conditions, monitoring node equipment health status, and the like.
The platform product service layer provides basic capability and an implementation framework of typical application, and developers can complete block chain implementation of business logic based on the basic capability and the characteristics of the superposed business. The application service layer provides the application service based on the block chain scheme for the business participants to use.
The design concept of the embodiment of the present application is described below.
With the development of internet technology and the increase of the amount of multimedia information, users will face huge amounts of multimedia information every day. To improve user experience, multimedia information applications typically recommend relevant multimedia information to a user based on the multimedia information currently viewed by the user.
When the multimedia information is applied to recommend the multimedia information to the user, since the multimedia information is simply screened, whether the user is interested in the recommended multimedia information cannot be estimated, the multimedia information which is not interested by the user may be pushed to the user, and troubles are caused to the user. And the multimedia information can adjust the multimedia information recommendation mode only after receiving the negative feedback of the user, and the comparison is lagged, for example, the video source is shielded aiming at the user.
For example, after a multimedia information application recommends a hunter video about eating betel nuts to a user, the user is very dislike the recommended video. After receiving the negative feedback of the user, the multimedia information application shields the video source for the user. However, the user still receives a similar hunter video, e.g., a hunter video about eating octopus, which makes the user experience poor.
Obviously, in the conventional technology, when determining to recommend multimedia information, only simple screening is performed on each piece of multimedia information, the accuracy of multimedia information recommendation is low, uninteresting multimedia information is usually recommended to a user, which leads to negative evaluation of the user, and accurate adjustment cannot be performed on multimedia information recommendation according to received negative feedback.
Considering that each multimedia information can be primarily screened and then screened again through the estimated user feedback, the embodiment of the application provides a scheme for recommending the multimedia information, in the scheme, through the user portrait of each sample user corresponding to the multimedia information label and the user feedback, training the deep neural network to obtain a feedback prediction model, and then, when recommending the multimedia information, after each multimedia information is preliminarily screened, the multimedia information labels of the screened candidate multimedia information and a correspondingly arranged feedback estimation model are obtained, the user portrait of the user to be recommended is input into the feedback estimation model, the feedback result of the user to be recommended to the candidate multimedia information is estimated, and only recommending interesting multimedia information to the user to be recommended according to the feedback result, and filtering the uninterested multimedia information of the user to be recommended.
To further illustrate the technical solutions provided by the embodiments of the present application, the following detailed description is made with reference to the accompanying drawings and the detailed description. Although the embodiments of the present application provide method steps as shown in the following embodiments or figures, more or fewer steps may be included in the method based on conventional or non-inventive efforts. In steps where no necessary causal relationship exists logically, the order of execution of the steps is not limited to that provided by the embodiments of the present application. The method can be executed in the order of the embodiments or the method shown in the drawings or in parallel in the actual process or the control device.
The method for recommending the multimedia information can be applied to multimedia information recommendation systems of various contents, such as watching point video recommendation, news recommendation, browser home page recommendation, micro-vision one-to-three recommendation and the like. Fig. 1a is a schematic diagram of a multimedia information recommendation system. The system comprises five parts of indexing, recalling, rough ranking, fine ranking and exhibition control, the multimedia information can be screened through the indexing, the recalling, the rough ranking and the fine ranking, and finally the screened multimedia information is recommended to a user through the exhibition control part.
The index is used for acquiring each multimedia information stored in the blocks in all the block chains. The input multimedia information is sequentially screened through recall, rough ranking and fine ranking, and each candidate multimedia information after screening can be obtained. And the exhibition control is used for presenting the finally screened specified amount of multimedia information to the user.
And each module in the recall, the rough ranking and the fine ranking is correspondingly provided with a corresponding screening condition so as to screen the multimedia information input by the previous layer through the set corresponding screening condition.
Wherein, the screening condition can comprise the following modes:
the first mode is as follows: and acquiring a multimedia information label of the multimedia information currently viewed by the user, and screening out the multimedia information corresponding to the multimedia information label.
The second way is: and screening the multimedia information according to user tags of the users, such as regions, languages, genders and the like.
For example, if the language of the user is Chinese, the video with the language of Chinese is screened out.
The third mode is as follows: and screening the multimedia information according to the operation records of the user, such as a retrieval record, a history record and the like.
For example, according to the user history of the user, if it is determined that the video played by the user is mainly a food program, each food video is screened out.
In the embodiment of the application, a multimedia information quality inspection module is added between the fine ranking and the exhibition control of the multimedia information recommendation system. The multimedia information quality inspection module is used for filtering out multimedia information which is not interesting to the user.
Fig. 1b is a schematic diagram of a multimedia information quality inspection module. The method comprises the following specific steps:
screening each multimedia information according to preset screening conditions through an indexing, recalling, coarse ranking and fine ranking module to obtain output candidate multimedia information, recommending the candidate multimedia information to a user to be recommended if a multimedia information label of the candidate multimedia information is not contained by a hot spot label set, otherwise, estimating user feedback of the user to the screened candidate multimedia information according to a user portrait of the user to be recommended and a pre-trained feedback estimation model, filtering the candidate multimedia information with a feedback result being negative feedback through the estimated feedback result, and recommending only the candidate multimedia information with the feedback result being non-negative feedback to the user to be recommended.
Therefore, the multimedia information is primarily screened, and then the multimedia information which is possibly sent out negative feedback by the user is screened again, so that the negative feedback rate of the multimedia information is reduced, the user experience is improved, and finally the retention rate, the daily average service life and other indexes of the multimedia information application are improved.
Fig. 1c shows an application scenario of multimedia information recommendation. The application scenario includes a plurality of terminal devices 110 and a server 130, and fig. 1c illustrates three terminal devices 110, which does not actually limit the number of terminal devices 110. The terminal device 110 is installed with a client 120 for viewing multimedia information. Client 120 and server 130 may communicate over a communication network. Terminal devices 110 are for example mobile phones, tablets and personal computers etc. The server 130 may be implemented by a single server or may be implemented by a plurality of servers. The server 130 may be implemented by a physical server or may be implemented by a virtual server.
In one possible application scenario, the servers 130 may be deployed in various regions in order to reduce the communication delay of the multimedia information recommendation, or different servers 130 may serve users respectively in order to balance the load. The plurality of servers 130 may implement sharing of multimedia information through a blockchain, and the plurality of servers 130 constitute a multimedia data sharing system. For example, the terminal device 110 corresponding to the user a is located at the location a and is in communication connection with the server 130, and the terminal device 110 corresponding to the user C, B is located at the location b and is in communication connection with the other server 130.
Each server 130 in the multimedia data sharing system has a node identifier corresponding to the server 130, and each server 130 in the multimedia data sharing system may store node identifiers of other servers in the data sharing system, so that the generated block is broadcast to other servers 130 in the multimedia data sharing system according to the node identifiers of other servers 130. Each server 130 may maintain a node identifier list as shown in the following table, and store the name of the server 130 and the node identifier in the node identifier list. The node identifier may be an IP (Internet Protocol) address and any other information that can be used to identify the node, and table 1 only illustrates the IP address as an example.
Table 1.
Server name Node identification
Node 1 119.115.151.174
Node 2 118.116.189.145
… …
Node N 119.123.789.258
In an embodiment, an administrator writes multimedia information into a multimedia data sharing system through a certain server 130, the multimedia data sharing system stores each piece of multimedia information, and any server 130 can obtain the multimedia information through a block of a block chain and send the determined candidate multimedia information to the client 120.
Optionally, the multimedia data sharing system may further store data in the multimedia recommendation and feedback estimation model training process, such as model parameters of the feedback estimation model, user images, and candidate multimedia information.
In the embodiment of the present application, before the multimedia information is recommended, a corresponding feedback prediction model is trained for each multimedia information tag, and it should be noted that, in the embodiment of the present application, only one multimedia information tag is trained for a feedback prediction model as an example, and based on the same principle, other multimedia information tags can obtain corresponding feedback prediction models.
Referring to fig. 2, a flowchart of an implementation of a method for training a feedback prediction model according to the present application is shown. The method comprises the following specific processes:
step 200: and the server acquires the user portrait of each sample user corresponding to the multimedia information label and user feedback.
Specifically, each sample user of the feedback estimation model corresponding to one multimedia information label is: and checking the multimedia information corresponding to the multimedia information label.
That is, the sample user is different when the multimedia information labels are different.
In one embodiment, if the application page viewed by the user includes the multimedia information corresponding to the multimedia information tag, the user is determined as the sample user of the multimedia information tag.
That is, if the multimedia information application presents the multimedia information to the user, the user is regarded as the sample user even if the user does not perform operations such as clicking and playing.
In practical applications, sample users can be classified into different categories according to actual requirements. The dividing standard of the sample category can be set according to actual requirements, and is not described herein again. In the embodiment of the present application, the sample users include at least a positive sample user and a negative sample user, and may further include a neutral sample user.
A positive sample user is a user that gives positive feedback. The negative sample user is a user who gives negative feedback. The neutral sample users are users other than the positive sample user and the non-negative sample user among the sample users.
The category of the user feedback can be determined according to the multimedia information viewing time and the user evaluation. In one embodiment, the user feedback includes positive feedback, negative feedback, and neutral feedback.
For example, negative feedback is feedback that the user evaluates as indicating dissatisfaction, positive feedback is feedback that the user evaluates as not indicating dissatisfaction, and any one or any combination of the following conditions are met: the viewing time is higher than the specified time, the viewing frequency is higher than the specified frequency, the ratio of the playing effective duration to the total duration is higher than the specified ratio, and the user evaluation represents satisfaction.
In practical application, the designated time, the designated frequency and the designated proportion may be set according to a practical application scene, and are not described herein again.
Therefore, the user with high positive feedback tendency is used as a positive sample, the user with high negative feedback tendency is used as a negative sample, and the user with fuzzy feedback tendency is used as a neutral sample, so that the feedback estimation model can be better fitted with the neutral sample through the differentiation of the positive and negative samples.
In an embodiment of the application, a user representation is a set comprising user tags and corresponding weights. The user tags at least comprise tags determined according to the multimedia information viewing operation. Wherein the higher the weight, the more important the user label is.
FIG. 3a is a diagram illustrating an example of a user representation. The user representation includes a user identification, a user label, and a weight.
Alternatively, the user representation may be represented in the following manner:
Ui＝[Wnorm(T(i，1))，Wnorm(T(i，2))，……Wnorm(T(i，n))]。
wherein, UiIs a user representation matrix with sequence number i, WnormThe normalized weight of the user label is T, the type of the user label is T, n is a positive integer, and the number of the user labels is N.
When T is used, it is noted that(i，t)In the absence of Wnorm(T(i，t)) 0. t is a positive integer. That is, when any user tag of the user does not exist, the corresponding weight is zero.
Optionally, the normalized weight may be obtained by using the following formula:
wherein i is the user number, WnormThe normalized weight of the user label, W, T, n, T0, n and T are positive integers, and represent the serial number of the user label.
In one embodiment, the user tag is determined according to operation data of the user for viewing the multimedia information, such as time for viewing the multimedia information and frequency for viewing the multimedia information, and classification of the user retrieval content.
Further, the user tags may also include tags obtained by desensitizing the user's base representation features determined based on user attributes.
Because the user attribute may expose the privacy of the user and is sensitive data, the user attribute is processed in a desensitization (such as linear transformation) mode, so as to improve the coverage of the user portrait on the premise of protecting the privacy security of the user, and further improve the accuracy of multimedia information recommendation.
Therefore, the user portrait of the user can be generated according to the user attribute of the user, the user behavior related to the multimedia information and the user interest.
Further, to improve the accuracy of the user representation, the user representation may be generated from data within a most recent specified time period. Thus, the obtained user portrait has certain hotspot property and can reflect the long-term preference of the user.
For example, a user representation is generated from data within 7 days prior to model training.
Step 201: and the server respectively counts the sample number of each user label in the user portrait according to the user portrait of each sample user.
Specifically, since the user labels contained in the user images of different sample users may be different, in order to improve the model fitting degree and reduce the model training data, the server respectively counts the number of samples of each user label, so as to filter the user labels in the subsequent steps.
In this way, the number of valid samples of the user tag can be counted.
Step 202: the server removes user tags corresponding to sample numbers lower than the specified sample number from the user representation of each sample user.
Specifically, the server executes the following steps for each user tag in the user representation:
the user tag is removed from the user representation when it is determined that the number of exemplars for the user tag is below a specified number of exemplars.
The specified sample number may be set according to an actual application scenario, and for example, the specified sample number is 10, which is not described herein again.
This is because the user profile includes a large number of user tags, for example, 5500 ten thousand user tags, and the enormous training data makes training of the model difficult. Furthermore, the number of valid user labels of a user is small, for example, 35 user labels are averaged, and obviously, after the user portrait of each sample user is converted into the portrait matrix, the portrait matrix may be too sparse, which may reduce the fitting effect during model training.
Therefore, in the embodiment of the application, each user label in the user portrait is screened, and the user labels with less occurrence times are abandoned, so that the model can be effectively trained.
Step 203: and the server adopts a dimension reduction processing algorithm to perform dimension reduction processing on the user portrait of each sample user to obtain the user portrait after the dimension reduction processing.
Optionally, the dimensionality reduction processing algorithm may employ a PCA algorithm.
The dimension reduction process reduces the dimension of the user image, thereby reducing the data amount of model training.
After the user labels are screened according to the number of the samples, the number of the user labels is still large, training parameters of the neural network are still in the tens of millions, the data volume of the network is huge, and meanwhile, the server is greatly burdened by deploying the network with the plurality of user labels to the server, so that the user portrait can be further reduced through a dimension reduction processing algorithm, and the data volume of model training is reduced.
Step 204: and the server trains the deep neural network according to the user portrait and the user feedback after the dimension reduction processing to obtain a feedback estimation model.
Specifically, the feedback prediction model is composed of a specified number of layers of fully-connected layers, a Batch-Normalization Layer (Batch-Normalization Layer), a random deactivation Layer (Drop-out) and an output Layer adopting an activation function. The feedback prediction model is obtained by training a Deep Neural Network (DNN) through user images and user feedback of sample users. The input of the feedback estimation model is a matrix formed by user portrait, and the output is the estimated feedback result.
In practical applications, the number of the designated layers may be set according to practical application scenarios, for example, the number of the designated layers may be 5, which is not limited herein.
Referring to fig. 3b, a schematic diagram of a feedback prediction model is shown, which includes a fully-connected Layer, a Batch-Normalization Layer, (Drop-out), and an output Layer using an activation function, where the fully-connected Layer includes a first Layer and an nth Layer … …, where n is a designated number of layers, and optionally n may be 5. For example, the full-link layer is denoted by D, the neuron number is denoted by [ ], the first full-link layer D1 ═ 2048, the second full-link layer D2 ═ 1024, the fifth full-link layer D5 ═ 16, and finally, the output layer outputs 3 feedback categories by a vector containing 3 elements.
Optionally, during model training, the feedback estimation model may be performed in the most popular neural network framework Keras currently, and specific model training is performed by using an Nvidia Titan-V display card.
Because adjacent user labels in the user portrait have no correlation, the converted user portrait matrix has no local correlation and can dig out relevant features, so that the feedback estimation model is not suitable for Neural networks (Recurrent Neural networks, RNNs) and CNNs and the like for capturing spatial information. Because the full-connection layer can find the cross relation among all input user labels, in the embodiment of the application, a feedback prediction model is obtained by adopting deep neural network training including the full-connection layer.
The parameter relation between all connection layers in the feedback estimation model is as follows:
A[D(m+1)]＝g[W[D(m+1)]A[Dm]+B[D(m+1)]]；
g(x)＝x；
wherein A is[D(m+1)]Is a parameter matrix of the m +1 th layer of the model, W[D(m+1)]Is the weight matrix of the m +1 th layer of the model, A[Dm]Is a parameter matrix of the m-th layer of the model, B[D(m+1)]The constant matrix of the (m + 1) th layer of the model is shown, m is the number of the fully-connected layers and is a positive number, and optionally, m can be 5.
It should be noted that, since the activation function of each fully-connected layer may return to 0 when performing forward propagation and backward propagation, the activation function used by each fully-connected layer is a linear layer (linear layer).
In the embodiment of the application, in order to prevent overfitting between certain layers of the feedback prediction model, an insertion normalization layer and a random deactivation layer are added after the second layer.
The Batch-Normalization layer normalizes the weights of the layer of neurons, so that the network is prevented from excessively concerning the values of the neurons with larger weights when the values of the network neurons are transmitted, the instability of the hidden layer is reduced, and overfitting is avoided.
Furthermore, when the Drop-out layer propagates forwards in the network, some neurons in the Drop-out layer are inactivated randomly, so that all the neurons can be trained fully, overfitting can be prevented through the method, and some neurons with higher dependence degree cannot be over-trained.
In the embodiment of the application, the model network of the feedback estimation model is a multi-classification task, so that Softmax is used for the final output layer activation function.
Alternatively, Softmax may use the following equation:
Si＝ei/∑jej；
wherein S isiThe feedback category with the index i is denoted by e, the probability is denoted by i, the index of each feedback category is denoted by j, and j is 1, 2, and 3.
The server depends on max (S)i) To ultimately determine the user feedback of the user to the multimedia information. That is, the most probable feedback category is determined as the feedback result.
For example, if the probability of negative feedback is 0.7, the estimated feedback result is determined to be negative feedback.
Referring to fig. 4, a flowchart of an implementation of a method for recommending multimedia information is provided. The method comprises the following specific processes:
step 400: and the server screens the acquired multimedia information according to preset screening conditions to obtain candidate multimedia information to be recommended.
Specifically, in practical application, the preset screening condition may be set according to a practical application scenario, which is not described herein again.
In one embodiment, the server obtains candidate multimedia information to be recommended after screening each multimedia information in the multimedia information database according to preset screening conditions through indexing, recalling, coarse ranking and fine ranking.
In the embodiment of the present application, only the user feedback of one candidate multimedia message is estimated for an example, in practical applications, the server usually obtains a specified number of candidate multimedia messages (e.g. 10 videos).
Therefore, massive multimedia information can be primarily screened through the preset screening conditions, so that in the subsequent steps, only a small amount of screened candidate multimedia information is filtered again through the feedback estimation model, the data volume of filtering processing can be reduced, and the accuracy of multimedia information recommendation can be improved.
Step 401: the server acquires the multimedia information labels of the candidate multimedia information.
Specifically, a corresponding multimedia information tag is set for each candidate multimedia information in advance, and the number of the multimedia information tags of each multimedia information may be set according to an actual application scenario.
For example, the multimedia information tag may be: games, animations, and gourmets, etc.
Step 402: the server determines whether the multimedia information tag is included in the acquired hotspot tag set, if so, step 403 is executed, otherwise, step 406 is executed.
Specifically, the server obtains a hot spot tag set, when the hot spot tag set does not include a multimedia information tag, the server recommends candidate multimedia information to the user to be recommended, and when the hot spot tag set includes a multimedia information tag, step 403 is executed, so that corresponding recommendation operation can be executed according to estimated user feedback.
Wherein, the hot spot label set is a set of a specified number of specified multimedia information labels.
In practical applications, the specified number may be set according to practical application scenarios, for example, the specified number may be 10, which is not limited herein.
Optionally, the designated multimedia information tag may be a tag of hot content obtained through big data statistics. For example, the designated multimedia information tag may be determined according to the number of times of retrieval, the number of times of browsing, the viewing time, and the like, which are counted.
Because when the user feedback of the multimedia information by the user is estimated, the feedback estimation model corresponding to the multimedia information tags of the multimedia information needs to be trained in advance, the number of the multimedia information tags is very large, if the corresponding feedback estimation model is trained for all the multimedia information tags, the cost is extremely high, which is obviously not in accordance with the actual situation, and the multimedia information tags of the hot content already cover most multimedia information, in the embodiment of the application, the hot tag set composed of the multimedia information tags of the hot content is obtained, and the corresponding feedback estimation model is trained only for each multimedia information tag included in the hot tag set.
Step 403: and the server acquires a feedback estimation model corresponding to the multimedia information label.
Specifically, before executing step 403, the server trains a corresponding feedback prediction model for each multimedia information tag in advance, and establishes a corresponding relationship between each multimedia information tag and each feedback prediction model.
Step 404: and the server predicts the feedback result of the user to be recommended to the candidate multimedia information according to the feedback prediction model and the acquired user portrait of the user to be recommended.
Step 405: and the server executes corresponding recommendation operation according to the feedback result.
Specifically, when step 405 is executed, the following two ways may be adopted:
the first mode is as follows: and when the feedback result represents positive feedback, the server recommends candidate multimedia information to the user to be recommended.
Specifically, the server determines whether the obtained feedback result represents that the number of the candidate multimedia information with positive feedback reaches the specified push number, if yes, a multimedia information list containing each candidate multimedia information is pushed to the multimedia information application of the user to be recommended, and if not, step 400 is executed.
Therefore, the multimedia information application can present the multimedia information list pushed by the server to the user to be recommended.
The second way is: and when the feedback result represents negative feedback, the server does not recommend candidate multimedia information to the user to be recommended.
Step 406: and the server recommends candidate multimedia information to the user to be recommended.
In the embodiment of the application, each acquired multimedia information is preliminarily screened, then the screened candidate multimedia information is filtered, and the multimedia information recommendation is optimized, so that the multimedia information recommendation accuracy is improved, the multimedia information recommendation which is not interested by a user is reduced, the negative feedback rate of the user is reduced, and the user experience is improved.
Next, a specific application scenario is adopted to test the effect of the trained feedback estimation model.
Suppose the multimedia information label glows for the king. The server obtains a user who performs negative feedback aiming at the video with the multimedia information label glory for the royal person in half a month of No. 5.10-5.25, namely a negative sample user, and a corresponding user portrait. In order to reduce the training data of the model, neutral sample users and positive sample users are extracted from the users who look at the video with the multimedia information labels glowing by the royal in half a month of No. 5.10-5.25. The number of positive sample users, negative sample users, and neutral sample users is: 6.4w, 8.12w and 12.5 w. And carrying out model training through user figures and user feedback of the positive sample user, the negative sample user and the neutral sample user to obtain a feedback estimation model.
And after the trained feedback estimation model is obtained, obtaining the feedback result of each user No. 5.26 estimated by the feedback estimation model aiming at the video, and calculating the prediction accuracy according to the actual feedback result and the estimated feedback result of each user.
Table 2.
Positive sample user Negative example user Neutral sample user
Prediction accuracy 94.45％ 85.42％ 77.34％
Referring to table 2, an example table of prediction accuracy is shown. The prediction accuracy of the positive sample user is as follows: the prediction result is positive feedback and the number of correct users is predicted, and the ratio of the actual feedback result to the total number of users with positive feedback is obtained.
The prediction accuracy of the negative sample user is as follows: the predicted result is negative feedback and predicts the correct number of users, compared to the actual feedback result which is the total number of users with negative feedback.
The prediction accuracy of the neutral sample user is the prediction result: non-positive feedback, non-negative feedback and predicting the correct number of users, and the ratio of the total number of users whose actual feedback result is non-positive feedback and non-negative feedback.
According to the statistical result, the feedback estimation model can effectively discover 85% of negative feedback users and intervene in time, so that the negative feedback generation can be greatly reduced, the user experience is improved, and the optimization of multimedia information recommendation is realized.
Based on the same inventive concept, the embodiment of the application also provides a device for recommending multimedia information, and as the principle of solving the problems of the device and the equipment is similar to that of a method for recommending multimedia information, the implementation of the device can be referred to the implementation of the method, and repeated parts are not described again.
Fig. 5 is a schematic structural diagram of an apparatus for recommending multimedia information according to an embodiment of the present application. An apparatus for multimedia information recommendation includes:
the screening unit 500 is configured to screen each acquired multimedia information according to a preset screening condition to obtain candidate multimedia information to be recommended;
a first obtaining unit 501, configured to obtain a multimedia information tag of candidate multimedia information to be recommended;
a second obtaining unit 502, configured to obtain a feedback estimation model set corresponding to a multimedia information tag and a user portrait of a user to be recommended, where the feedback estimation model corresponding to one multimedia information tag is used to predict, according to the user portrait of the user, feedback of the user to multimedia information corresponding to the multimedia information tag;
the estimating unit 503 is configured to estimate a feedback result of the user to be recommended for the candidate multimedia information according to the feedback estimation model and the user portrait;
and the recommending unit 504 is configured to not recommend the candidate multimedia information to the user to be recommended when the feedback result represents negative feedback, and recommend the candidate multimedia information to the user to be recommended when the feedback result represents non-negative feedback.
Preferably, the first obtaining unit 501 is further configured to:
acquiring a hotspot tag set, wherein the hotspot tag set is a set of appointed multimedia information tags with an appointed number;
when the hotspot tag set does not contain the multimedia information tag, recommending candidate multimedia information to the user to be recommended;
and when the hot spot label set comprises the multimedia information labels, executing the step of acquiring the feedback estimation model set corresponding to the multimedia information labels.
Preferably, the feedback estimation model consists of a specified number of full connection layers, an insertion normalization layer, a random inactivation layer and an output layer adopting an activation function, and is obtained by training the deep neural network through user images and user feedback of sample users;
each sample user of the feedback estimation model corresponding to one multimedia information label is a user for viewing the multimedia information corresponding to the multimedia information label;
a user representation is a collection containing user labels and corresponding weights;
the user tags at least comprise tags determined according to the multimedia information viewing operation;
the user feedback is determined based on the multimedia information viewing time and the user rating.
Preferably, the user label also comprises a label obtained after desensitization processing is carried out on the basic portrait characteristics of the user;
the base representation features are determined based on user attributes.
Preferably, the feedback estimation model corresponding to the multimedia information label is obtained by training in the following way:
acquiring user portraits and user feedback of sample users corresponding to the multimedia information labels;
respectively counting the sample number of each user label in each user portrait according to the user portrait of each sample user;
removing user tags of which the corresponding sample quantity is lower than the specified sample quantity from the user portrait of each sample user;
performing dimension reduction processing on the user portrait of each sample user by adopting a dimension reduction processing algorithm to obtain each dimension-reduced user portrait;
and training the deep neural network according to the user portrait and the user feedback after the dimension reduction processing to obtain a feedback estimation model.
Preferably, each multimedia message is obtained from each block in the block chain.
In the method, the device, the equipment and the medium for recommending the multimedia information, the obtained multimedia information is screened according to the preset screening condition, and candidate multimedia information to be recommended is obtained; and acquiring a multimedia information label of the candidate multimedia information to be recommended, and predicting a feedback result of the candidate multimedia information of the user to be recommended through a feedback prediction model and a user portrait which are set corresponding to the multimedia information label. And recommending the candidate multimedia information to the user to be recommended only when the feedback result represents non-negative feedback, and otherwise, filtering the candidate multimedia information. Therefore, on one hand, the multimedia information is preliminarily screened through the preset screening condition, on the other hand, the feedback of the multimedia information of the user is estimated through the feedback estimation model, so that the recommended multimedia information is screened again according to the estimated feedback result, the multimedia information recommendation is optimized, the precision of the multimedia information recommendation is improved, the multimedia information recommendation which is not interested by the user is reduced, the negative feedback rate of the user is reduced, and the user experience is improved.
Fig. 6 is a schematic structural diagram of a control device. Based on the same technical concept, the embodiment of the present application further provides a control device, which may include a memory 601 and a processor 602.
A memory 601 for storing computer programs executed by the processor 602. The memory 601 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the blockchain node, and the like. The processor 602 may be a Central Processing Unit (CPU), a digital processing unit, or the like. The specific connection medium between the memory 601 and the processor 602 is not limited in the embodiments of the present application. In the embodiment of the present application, the memory 601 and the processor 602 are connected by a bus 603 in fig. 6, the bus 603 is represented by a thick line in fig. 6, and the connection manner between other components is merely for illustrative purposes and is not limited thereto. The bus 603 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in FIG. 6, but this is not intended to represent only one bus or type of bus.
The memory 601 may be a volatile memory (volatile memory), such as a random-access memory (RAM); the memory 601 may also be a non-volatile memory (non-volatile) such as a read-only memory (rom), a flash memory (flash memory), a hard disk (HDD) or a solid-state drive (SSD), or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory 601 may be a combination of the above memories.
A processor 602 for executing the method of multimedia information recommendation provided by the embodiment shown in fig. 4 when calling the computer program stored in the memory 601.
The embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the method for recommending multimedia information in any of the above method embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a general hardware platform, and certainly can also be implemented by hardware. Based on such understanding, the above technical solutions substantially or partially contributing to the related art may be embodied in the form of a software product, which may be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions for enabling a control device (which may be a personal computer, a server, or a network device, etc.) to execute the methods of the various embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims (14)

1. A method for multimedia information recommendation, comprising:
screening the acquired multimedia information according to preset screening conditions to obtain candidate multimedia information to be recommended;
acquiring a multimedia information tag of candidate multimedia information to be recommended;
acquiring a feedback estimation model set corresponding to the multimedia information tag and a user portrait of a user to be recommended, wherein the feedback estimation model corresponding to the multimedia information tag is used for predicting the feedback of the user to the multimedia information corresponding to the multimedia information tag according to the user portrait of the user;
according to the feedback estimation model and the user portrait, estimating a feedback result of the user to be recommended to the candidate multimedia information;
when the feedback result represents negative feedback, the candidate multimedia information is not recommended to the user to be recommended;
and when the feedback result represents non-negative feedback, recommending the candidate multimedia information to the user to be recommended.
2. The method of claim 1, prior to obtaining the feedback prediction model corresponding to the multimedia information tag setting, further comprising:
acquiring a hotspot tag set, wherein the hotspot tag set is a set of appointed multimedia information tags with an appointed number;
when the hot spot label set does not contain the multimedia information label, recommending the candidate multimedia information to the user to be recommended;
and when the hot spot label set comprises the multimedia information label, executing the step of acquiring the feedback estimation model corresponding to the multimedia information label.
3. The method of claim 1, wherein the feedback prediction model consists of a specified number of fully-connected layers, an interpolated normalization layer, a random deactivation layer, and an output layer using an activation function, and is obtained by training a deep neural network through user images and user feedback of each sample user;
each sample user of the feedback estimation model corresponding to one multimedia information label is a user for viewing the multimedia information corresponding to the multimedia information label;
the user representation is a collection comprising user labels and corresponding weights;
the user tags at least comprise tags determined according to multimedia information viewing operation;
the user feedback is determined based on the multimedia information viewing time and the user rating.
4. The method of claim 3, wherein the user tags further comprise tags obtained by desensitizing the base representation features of the user;
the base portrait features are determined based on user attributes.
5. The method as claimed in claim 3 or 4, wherein the feedback prediction model corresponding to the multimedia information tag is obtained by training:
acquiring user portrait and user feedback of each sample user corresponding to the multimedia information label;
respectively counting the sample number of each user label in each user portrait according to the user portrait of each sample user;
removing user tags of which the corresponding sample quantity is lower than the specified sample quantity from the user portrait of each sample user;
performing dimension reduction processing on the user portrait of each sample user by adopting a dimension reduction processing algorithm to obtain each dimension-reduced user portrait;
and training the deep neural network according to the user portrait and the user feedback after the dimension reduction processing to obtain a feedback estimation model.
6. The method of any of claims 1-4, wherein the multimedia information is obtained from blocks in a block chain.
7. An apparatus for multimedia information recommendation, comprising:
the screening unit is used for screening the acquired multimedia information according to preset screening conditions to obtain candidate multimedia information to be recommended;
the first obtaining unit is used for obtaining a multimedia information label of candidate multimedia information to be recommended;
the second obtaining unit is used for obtaining a feedback estimation model set corresponding to the multimedia information tag and a user portrait of a user to be recommended, and the feedback estimation model corresponding to one multimedia information tag is used for predicting the feedback of the user to the multimedia information corresponding to the multimedia information tag according to the user portrait of the user;
the pre-estimation unit is used for pre-estimating a feedback result of the user to be recommended to the candidate multimedia information according to the feedback pre-estimation model and the user portrait;
and the recommending unit is used for not recommending the candidate multimedia information to the user to be recommended when the feedback result represents negative feedback, and recommending the candidate multimedia information to the user to be recommended when the feedback result represents non-negative feedback.
8. The apparatus of claim 7, wherein the first obtaining unit is further configured to:
acquiring a hotspot tag set, wherein the hotspot tag set is a set of appointed multimedia information tags with an appointed number;
when the hot spot label set does not contain the multimedia information label, recommending the candidate multimedia information to the user to be recommended;
and when the hot spot label set comprises the multimedia information label, executing the step of acquiring the feedback estimation model corresponding to the multimedia information label.
9. The apparatus of claim 7, wherein the feedback prediction model is composed of a specified number of fully-connected layers, an interpolated normalization layer, a random deactivation layer, and an output layer using an activation function, and is obtained by training a deep neural network through user images and user feedback of each sample user;
each sample user of the feedback estimation model corresponding to one multimedia information label is a user for viewing the multimedia information corresponding to the multimedia information label;
the user representation is a collection comprising user labels and corresponding weights;
the user tags at least comprise tags determined according to multimedia information viewing operation;
the user feedback is determined based on the multimedia information viewing time and the user rating.
10. The apparatus of claim 9, wherein the user tags further comprise tags obtained by desensitizing the base representation characteristics of the user;
the base portrait features are determined based on user attributes.
11. The apparatus according to claim 9 or 10, wherein the feedback prediction model corresponding to the multimedia information tag is obtained by training:
acquiring user portrait and user feedback of each sample user corresponding to the multimedia information label;
respectively counting the sample number of each user label in each user portrait according to the user portrait of each sample user;
removing user tags of which the corresponding sample quantity is lower than the specified sample quantity from the user portrait of each sample user;
performing dimension reduction processing on the user portrait of each sample user by adopting a dimension reduction processing algorithm to obtain each dimension-reduced user portrait;
and training the deep neural network according to the user portrait and the user feedback after the dimension reduction processing to obtain a feedback estimation model.
12. The apparatus of any of claims 7-10, wherein the multimedia information is obtained from blocks in a block chain.
13. A control apparatus, characterized by comprising:
at least one memory for storing program instructions;
at least one processor for calling program instructions stored in said memory and for executing the steps of the method of any of the preceding claims 1 to 6 in accordance with the program instructions obtained.
14. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.
CN201910857435.0A 2019-09-11 2019-09-11 Method, device, equipment and medium for recommending multimedia information Active CN110598016B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910857435.0A CN110598016B (en) 2019-09-11 2019-09-11 Method, device, equipment and medium for recommending multimedia information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910857435.0A CN110598016B (en) 2019-09-11 2019-09-11 Method, device, equipment and medium for recommending multimedia information

Publications (2)

Publication Number Publication Date
CN110598016A true CN110598016A (en) 2019-12-20
CN110598016B CN110598016B (en) 2021-08-17

Family

ID=68858797

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910857435.0A Active CN110598016B (en) 2019-09-11 2019-09-11 Method, device, equipment and medium for recommending multimedia information

Country Status (1)

Country Link
CN (1) CN110598016B (en)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111191055A (en) * 2020-01-02 2020-05-22 广州虎牙科技有限公司 Multimedia data processing method and device, computer equipment and storage medium
CN111310053A (en) * 2020-03-03 2020-06-19 上海喜马拉雅科技有限公司 Information recommendation method, device, equipment and storage medium
CN111339327A (en) * 2020-02-20 2020-06-26 北京达佳互联信息技术有限公司 Work recommendation method and device, server and storage medium
CN111784301A (en) * 2020-07-02 2020-10-16 中国银行股份有限公司 User portrait construction method and device, storage medium and electronic equipment
CN111898031A (en) * 2020-08-14 2020-11-06 腾讯科技（深圳）有限公司 Method and device for obtaining user portrait
CN112148994A (en) * 2020-10-23 2020-12-29 腾讯科技（深圳）有限公司 Information push effect evaluation method and device, electronic equipment and storage medium
CN112559880A (en) * 2020-12-24 2021-03-26 百果园技术(新加坡)有限公司 Information recommendation management method, system, equipment and storage medium
CN112685641A (en) * 2020-12-31 2021-04-20 五八有限公司 Information processing method and device
CN113011912A (en) * 2021-01-22 2021-06-22 腾讯科技（深圳）有限公司 Media information processing method, device, electronic equipment and storage medium
CN113032686A (en) * 2019-12-25 2021-06-25 北京达佳互联信息技术有限公司 Content recommendation method, device, storage medium and server
CN113190758A (en) * 2021-05-21 2021-07-30 聚好看科技股份有限公司 Server and media asset recommendation method
CN113407772A (en) * 2020-03-16 2021-09-17 北京达佳互联信息技术有限公司 Video recommendation model generation method, video recommendation method and device
CN113434753A (en) * 2020-03-23 2021-09-24 北京达佳互联信息技术有限公司 Information pushing method and device, electronic equipment and storage medium
CN113449175A (en) * 2020-03-24 2021-09-28 北京沃东天骏信息技术有限公司 Hot data recommendation method and device
CN113516565A (en) * 2021-04-08 2021-10-19 国家电网有限公司 Intelligent alarm processing method and device for power monitoring system based on knowledge base
CN113542893A (en) * 2020-04-14 2021-10-22 北京达佳互联信息技术有限公司 Method and device for acquiring evaluation information of works, and video screening method and device
CN113988483A (en) * 2021-12-23 2022-01-28 支付宝(杭州)信息技术有限公司 Risk operation behavior control method, risk operation behavior model training method and electronic equipment
CN114419501A (en) * 2022-01-11 2022-04-29 平安普惠企业管理有限公司 Video recommendation method and device, computer equipment and storage medium
CN114531614A (en) * 2022-02-17 2022-05-24 北京字跳网络技术有限公司 Video playing method, video playing device, electronic equipment, storage medium and program product
CN113032686B (en) * 2019-12-25 2024-05-28 北京达佳互联信息技术有限公司 Content recommendation method, device, storage medium and server

Citations (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030030004A1 (en) * 2001-01-31 2003-02-13 General Electric Company Shared memory control between detector framing node and processor
US20060271691A1 (en) * 2005-05-23 2006-11-30 Picateers, Inc. System and method for collaborative image selection
US20090100094A1 (en) * 2007-10-15 2009-04-16 Xavier Verdaguer Recommendation system and method for multimedia content
CN101489107A (en) * 2009-01-21 2009-07-22 华东师范大学 Collaborative filtering recommendation method based on population attribute keyword vector
US20100332443A1 (en) * 2009-06-25 2010-12-30 Daniel Gartenberg Cyclical Behavior Modification
US20120297038A1 (en) * 2011-05-16 2012-11-22 Microsoft Corporation Recommendations for Social Network Based on Low-Rank Matrix Recovery
CN102957949A (en) * 2012-05-18 2013-03-06 华东师范大学 Device and method for recommending video to user
CN103324742A (en) * 2013-06-28 2013-09-25 百度在线网络技术（北京）有限公司 Method and equipment for recommending keywords
US20130322685A1 (en) * 2012-06-04 2013-12-05 Ebay Inc. System and method for providing an interactive shopping experience via webcam
CN103559193A (en) * 2013-09-10 2014-02-05 浙江大学 Topic modeling method based on selected cell
CN104156390A (en) * 2014-07-07 2014-11-19 乐视网信息技术（北京）股份有限公司 Comment recommendation method and system
CN104166713A (en) * 2014-08-14 2014-11-26 百度在线网络技术（北京）有限公司 Network service recommending method and device
CN104216931A (en) * 2013-05-29 2014-12-17 酷盛(天津）科技有限公司 Real-time recommending system and method
CN104965890A (en) * 2015-06-17 2015-10-07 深圳市腾讯计算机系统有限公司 Advertisement recommendation method and apparatus
CN105068661A (en) * 2015-09-07 2015-11-18 百度在线网络技术（北京）有限公司 Man-machine interaction method and system based on artificial intelligence
CN105183731A (en) * 2014-06-04 2015-12-23 腾讯科技（深圳）有限公司 Method, device, and system for generating recommended information
CN105574110A (en) * 2015-12-14 2016-05-11 北京奇虎科技有限公司 Intelligent game recommending method and device
CN105915438A (en) * 2016-04-15 2016-08-31 北京奇虎科技有限公司 Message pushing method, apparatus, and system
CN105930425A (en) * 2016-04-18 2016-09-07 乐视控股（北京）有限公司 Personalized video recommendation method and apparatus
CN106326351A (en) * 2016-08-05 2017-01-11 天津大学 Recommendation system cold start solving method based on user feedback
CN106599022A (en) * 2016-11-01 2017-04-26 中山大学 User portrait forming method based on user access data
CN107463701A (en) * 2017-08-15 2017-12-12 北京百度网讯科技有限公司 Method and apparatus based on artificial intelligence pushed information stream
CN107992531A (en) * 2017-11-21 2018-05-04 吉浦斯信息咨询（深圳）有限公司 News personalization intelligent recommendation method and system based on deep learning
CN108257053A (en) * 2018-01-19 2018-07-06 通辽霍林河坑口发电有限责任公司 Safe actual effect approaches to IM and device
CN108307207A (en) * 2017-12-28 2018-07-20 北京达佳互联信息技术有限公司 A kind of video pushing method and device
CN110162698A (en) * 2019-04-18 2019-08-23 腾讯科技（深圳）有限公司 A kind of user's representation data processing method, device and storage medium
CN111222030A (en) * 2018-11-27 2020-06-02 阿里巴巴集团控股有限公司 Information recommendation method and device and electronic equipment

Patent Citations (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030030004A1 (en) * 2001-01-31 2003-02-13 General Electric Company Shared memory control between detector framing node and processor
US20060271691A1 (en) * 2005-05-23 2006-11-30 Picateers, Inc. System and method for collaborative image selection
US20090100094A1 (en) * 2007-10-15 2009-04-16 Xavier Verdaguer Recommendation system and method for multimedia content
CN101489107A (en) * 2009-01-21 2009-07-22 华东师范大学 Collaborative filtering recommendation method based on population attribute keyword vector
US20100332443A1 (en) * 2009-06-25 2010-12-30 Daniel Gartenberg Cyclical Behavior Modification
US20120297038A1 (en) * 2011-05-16 2012-11-22 Microsoft Corporation Recommendations for Social Network Based on Low-Rank Matrix Recovery
CN102957949A (en) * 2012-05-18 2013-03-06 华东师范大学 Device and method for recommending video to user
US20130322685A1 (en) * 2012-06-04 2013-12-05 Ebay Inc. System and method for providing an interactive shopping experience via webcam
CN104216931A (en) * 2013-05-29 2014-12-17 酷盛(天津）科技有限公司 Real-time recommending system and method
CN103324742A (en) * 2013-06-28 2013-09-25 百度在线网络技术（北京）有限公司 Method and equipment for recommending keywords
CN103559193A (en) * 2013-09-10 2014-02-05 浙江大学 Topic modeling method based on selected cell
CN105183731A (en) * 2014-06-04 2015-12-23 腾讯科技（深圳）有限公司 Method, device, and system for generating recommended information
CN104156390A (en) * 2014-07-07 2014-11-19 乐视网信息技术（北京）股份有限公司 Comment recommendation method and system
CN104166713A (en) * 2014-08-14 2014-11-26 百度在线网络技术（北京）有限公司 Network service recommending method and device
CN104965890A (en) * 2015-06-17 2015-10-07 深圳市腾讯计算机系统有限公司 Advertisement recommendation method and apparatus
CN105068661A (en) * 2015-09-07 2015-11-18 百度在线网络技术（北京）有限公司 Man-machine interaction method and system based on artificial intelligence
CN105574110A (en) * 2015-12-14 2016-05-11 北京奇虎科技有限公司 Intelligent game recommending method and device
CN105915438A (en) * 2016-04-15 2016-08-31 北京奇虎科技有限公司 Message pushing method, apparatus, and system
CN105930425A (en) * 2016-04-18 2016-09-07 乐视控股（北京）有限公司 Personalized video recommendation method and apparatus
CN106326351A (en) * 2016-08-05 2017-01-11 天津大学 Recommendation system cold start solving method based on user feedback
CN106599022A (en) * 2016-11-01 2017-04-26 中山大学 User portrait forming method based on user access data
CN107463701A (en) * 2017-08-15 2017-12-12 北京百度网讯科技有限公司 Method and apparatus based on artificial intelligence pushed information stream
CN107992531A (en) * 2017-11-21 2018-05-04 吉浦斯信息咨询（深圳）有限公司 News personalization intelligent recommendation method and system based on deep learning
CN108307207A (en) * 2017-12-28 2018-07-20 北京达佳互联信息技术有限公司 A kind of video pushing method and device
CN108257053A (en) * 2018-01-19 2018-07-06 通辽霍林河坑口发电有限责任公司 Safe actual effect approaches to IM and device
CN111222030A (en) * 2018-11-27 2020-06-02 阿里巴巴集团控股有限公司 Information recommendation method and device and electronic equipment
CN110162698A (en) * 2019-04-18 2019-08-23 腾讯科技（深圳）有限公司 A kind of user's representation data processing method, device and storage medium

Cited By (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113032686B (en) * 2019-12-25 2024-05-28 北京达佳互联信息技术有限公司 Content recommendation method, device, storage medium and server
CN113032686A (en) * 2019-12-25 2021-06-25 北京达佳互联信息技术有限公司 Content recommendation method, device, storage medium and server
CN111191055A (en) * 2020-01-02 2020-05-22 广州虎牙科技有限公司 Multimedia data processing method and device, computer equipment and storage medium
CN111339327A (en) * 2020-02-20 2020-06-26 北京达佳互联信息技术有限公司 Work recommendation method and device, server and storage medium
CN111339327B (en) * 2020-02-20 2024-05-07 北京达佳互联信息技术有限公司 Work recommendation method and device, server and storage medium
CN111310053A (en) * 2020-03-03 2020-06-19 上海喜马拉雅科技有限公司 Information recommendation method, device, equipment and storage medium
CN113407772A (en) * 2020-03-16 2021-09-17 北京达佳互联信息技术有限公司 Video recommendation model generation method, video recommendation method and device
CN113407772B (en) * 2020-03-16 2023-12-26 北京达佳互联信息技术有限公司 Video recommendation model generation method, video recommendation method and device
CN113434753B (en) * 2020-03-23 2024-03-19 北京达佳互联信息技术有限公司 Information pushing method and device, electronic equipment and storage medium
CN113434753A (en) * 2020-03-23 2021-09-24 北京达佳互联信息技术有限公司 Information pushing method and device, electronic equipment and storage medium
CN113449175A (en) * 2020-03-24 2021-09-28 北京沃东天骏信息技术有限公司 Hot data recommendation method and device
CN113542893A (en) * 2020-04-14 2021-10-22 北京达佳互联信息技术有限公司 Method and device for acquiring evaluation information of works, and video screening method and device
CN113542893B (en) * 2020-04-14 2023-08-18 北京达佳互联信息技术有限公司 Method and device for acquiring work evaluation information, and video screening method and device
CN111784301A (en) * 2020-07-02 2020-10-16 中国银行股份有限公司 User portrait construction method and device, storage medium and electronic equipment
CN111898031B (en) * 2020-08-14 2024-04-05 腾讯科技（深圳）有限公司 Method and device for obtaining user portrait
CN111898031A (en) * 2020-08-14 2020-11-06 腾讯科技（深圳）有限公司 Method and device for obtaining user portrait
CN112148994B (en) * 2020-10-23 2024-04-19 腾讯科技（深圳）有限公司 Information push effect evaluation method and device, electronic equipment and storage medium
CN112148994A (en) * 2020-10-23 2020-12-29 腾讯科技（深圳）有限公司 Information push effect evaluation method and device, electronic equipment and storage medium
CN112559880A (en) * 2020-12-24 2021-03-26 百果园技术(新加坡)有限公司 Information recommendation management method, system, equipment and storage medium
CN112685641A (en) * 2020-12-31 2021-04-20 五八有限公司 Information processing method and device
CN113011912B (en) * 2021-01-22 2023-07-25 腾讯科技（深圳）有限公司 Media information processing method, device, electronic equipment and storage medium
CN113011912A (en) * 2021-01-22 2021-06-22 腾讯科技（深圳）有限公司 Media information processing method, device, electronic equipment and storage medium
CN113516565A (en) * 2021-04-08 2021-10-19 国家电网有限公司 Intelligent alarm processing method and device for power monitoring system based on knowledge base
CN113190758B (en) * 2021-05-21 2023-01-20 聚好看科技股份有限公司 Server and media asset recommendation method
CN113190758A (en) * 2021-05-21 2021-07-30 聚好看科技股份有限公司 Server and media asset recommendation method
CN113988483B (en) * 2021-12-23 2022-04-29 支付宝(杭州)信息技术有限公司 Risk operation behavior control method, risk operation behavior model training method and electronic equipment
CN113988483A (en) * 2021-12-23 2022-01-28 支付宝(杭州)信息技术有限公司 Risk operation behavior control method, risk operation behavior model training method and electronic equipment
CN114419501A (en) * 2022-01-11 2022-04-29 平安普惠企业管理有限公司 Video recommendation method and device, computer equipment and storage medium
CN114531614A (en) * 2022-02-17 2022-05-24 北京字跳网络技术有限公司 Video playing method, video playing device, electronic equipment, storage medium and program product
CN114531614B (en) * 2022-02-17 2023-09-15 北京字跳网络技术有限公司 Video playing method, device, electronic equipment and storage medium

Also Published As

Publication number Publication date
CN110598016B (en) 2021-08-17

Similar Documents

Publication Publication Date Title
CN110598016B (en) Method, device, equipment and medium for recommending multimedia information
Huang et al. Data poisoning attacks to deep learning based recommender systems
Pacheco et al. Uncovering coordinated networks on social media
CN109544197B (en) User loss prediction method and device
Xu et al. Course video recommendation with multimodal information in online learning platforms: A deep learning framework
CN111522724B (en) Method and device for determining abnormal account number, server and storage medium
CN113283948B (en) Generation method, device, equipment and readable medium of prediction model
CN111553744A (en) Federal product recommendation method, device, equipment and computer storage medium
CN113761359B (en) Data packet recommendation method, device, electronic equipment and storage medium
CN112257841A (en) Data processing method, device and equipment in graph neural network and storage medium
WO2023231542A1 (en) Representation information determination method and apparatus, and device and storage medium
CN111931043B (en) Recommending method and system for science and technology resources
Cheong et al. Interpretable stock anomaly detection based on spatio-temporal relation networks with genetic algorithm
Kumar A novel weakest t-norm based fuzzy fault tree analysis through qualitative data processing and its application in system reliability evaluation
Fang et al. Subjectivity grouping: learning from users' rating behavior
CN115730125A (en) Object identification method and device, computer equipment and storage medium
US20230316106A1 (en) Method and apparatus for training content recommendation model, device, and storage medium
CA2651165A1 (en) Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising
CN113487389A (en) Information recommendation method and device
CN112541669A (en) Risk identification method, system and device
Chung et al. A temporal graph framework for intelligence extraction in social media networks
US11010442B2 (en) Systems and methods for intelligence delivery
CN114820085B (en) User screening method, related device and storage medium
Kalameyets Algorithms and techniques for bot detection in social networks
Wu et al. A privacy-preserving student status monitoring system

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20221118

Address after: 1402, Floor 14, Block A, Haina Baichuan Headquarters Building, No. 6, Baoxing Road, Haibin Community, Xin'an Street, Bao'an District, Shenzhen, Guangdong 518133

Patentee after: Shenzhen Yayue Technology Co.,Ltd.

Address before: 35th floor, Tencent building, Keji Zhongyi Road, high tech Zone, Nanshan District, Shenzhen City, Guangdong Province

Patentee before: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd.