WO2022262561A1 - 多媒体资源的处理方法、装置、设备及存储介质 - Google Patents

多媒体资源的处理方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2022262561A1
WO2022262561A1 PCT/CN2022/095887 CN2022095887W WO2022262561A1 WO 2022262561 A1 WO2022262561 A1 WO 2022262561A1 CN 2022095887 W CN2022095887 W CN 2022095887W WO 2022262561 A1 WO2022262561 A1 WO 2022262561A1
Authority
WO
WIPO (PCT)
Prior art keywords
relationship
user
target user
interaction
feature information
Prior art date
Application number
PCT/CN2022/095887
Other languages
English (en)
French (fr)
Inventor
陈昊
Original Assignee
腾讯科技（深圳）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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2022262561A1 publication Critical patent/WO2022262561A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/45—Clustering; Classification
    • 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
    • G06F16/43—Querying
    • G06F16/435—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—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
    • G06F16/483—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present application relates to the field of computer technology, and in particular to a multimedia resource processing method, device, equipment, and computer-readable storage medium.
  • multimedia resources With the development of computer technology, massive multimedia resources emerge in the network.
  • most recommendation algorithms for multimedia resources are based on the user's historical browsing records to mine (discover) the user's points of interest, and then recommend multimedia resources that may be of interest to the user.
  • the existing recommendation algorithms are highly dependent on the user's historical browsing records.
  • the user's actual browsing behavior is small, such as the current user is a new user, or the platform's business is in its infancy
  • the user's browsing records that can be obtained are very limited, making multimedia resources recommended based on existing recommendation algorithms different from current
  • the matching degree of multimedia resources that users are actually interested in is low. Therefore, in this scenario, the recommendation accuracy of existing recommendation algorithms is low.
  • Embodiments of the present application provide a multimedia resource processing method, device, device, and storage medium, which can effectively improve recommendation accuracy in the absence of historical browsing data.
  • an embodiment of the present application provides a method for processing multimedia resources, executed by an electronic device, including:
  • an interaction vector of the target user and a set of interaction vectors of adjacent users are obtained, the interaction vector set of adjacent users includes an interaction vector of a first adjacent user of the target user, the The relationship between the target user and the first adjacent user is divided into K first relationship types, where K is a positive integer;
  • the social feature information of the target user is obtained, and the social feature information is based on the target user in the K first relationship types , determined by the relationship characteristic information corresponding to each first relationship type;
  • the embodiment of the present application also provides a multimedia resource processing device, including:
  • An acquisition unit configured to acquire interaction data between terminal devices of multiple users; construct a social relationship network diagram according to the interaction data; acquire an interaction vector of a target user and an interaction of adjacent users according to the social relationship network diagram A set of vectors, the interaction vector set of adjacent users includes the interaction vector of the first adjacent user of the target user, and the relationship between the target user and the first adjacent user is divided into Kth A relationship type, K is a positive integer;
  • a processing unit configured to obtain social feature information of the target user according to the interaction vector of the target user and the set of interaction vectors of the adjacent users, the social feature information is based on the target user's interaction among the K In the first relationship type, the relationship characteristic information corresponding to each first relationship type is determined; acquire a multimedia resource set, and display the multimedia on the terminal device of the target user according to the social characteristic information of the target user The first multimedia resource in the collection of resources.
  • an embodiment of the present application also provides an electronic device, including: a storage device and a processor; a computer program is stored in the storage device; and the processor executes the computer program to implement the above-mentioned method for processing multimedia resources.
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for processing multimedia resources is realized.
  • the present application provides a computer program product or computer program
  • the computer program product or computer program includes computer instructions
  • the computer instructions are stored in a computer-readable storage medium
  • the processor of the computer device reads the The computer instruction is read, and the processor executes the computer instruction, so that the computer device executes the above-mentioned method for processing multimedia resources.
  • FIG. 1 is a processing scene diagram of a multimedia resource provided by an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a method for processing multimedia resources provided by an embodiment of the present application
  • FIG. 3 is a schematic flowchart of another method for processing multimedia resources provided by an embodiment of the present application.
  • FIG. 4a is a schematic diagram of a user social relationship network diagram provided by an embodiment of the present application.
  • FIG. 4b is a schematic diagram of a graph convolution model based on a social relationship network graph provided in an embodiment of the present application
  • FIG. 5 is a schematic structural diagram of an apparatus for processing multimedia resources provided by an embodiment of the present application.
  • FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • AI Artificial Intelligence
  • Machine Learning Machine Learning
  • ML Machine Learning
  • AI is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
  • artificial intelligence is a comprehensive technology of computer science; it mainly produces a new electronic device that can respond in a similar way to human intelligence by understanding the essence of intelligence, making electronic devices capable of perception, reasoning and decision-making and other functions.
  • the electronic device provided by the embodiment of the present application can recommend multimedia resources to a target user with less browsing behavior based on a relationship type between the target user and adjacent users of the target user.
  • AI technology is a comprehensive subject that involves a wide range of fields, including both hardware-level technology and software-level technology.
  • AI basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, processing technology for large applications, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes several major directions such as computer vision technology, speech processing technology, natural language processing (Nature Language processing, NLP) technology, and machine learning/deep learning.
  • one or more of the aforementioned AI software technologies will be involved when converting the interaction behavior between users into edge information. For example, when extracting the features of (short) videos sent between users, computer vision technology will be involved; when extracting the features of voice sent between users, voice processing technology will be involved; when extracting features of text information sent between users , will involve natural language processing technology.
  • computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, optical character recognition (Optical Character Recognition, OCR), video processing, video semantic understanding, video content/behavior recognition, 3D object reconstruction, 3D technology , virtual reality, augmented reality, simultaneous positioning and map construction technologies, as well as common biometric recognition technologies such as face recognition and fingerprint recognition.
  • OCR optical Character Recognition
  • video processing video semantic understanding, video content/behavior recognition
  • 3D object reconstruction 3D technology
  • virtual reality augmented reality
  • simultaneous positioning and map construction technologies as well as common biometric recognition technologies such as face recognition and fingerprint recognition.
  • Speech processing technology includes automatic speech recognition technology (Automatic Speech Recognition, ASR) and speech synthesis technology (Text To Speech, TTS) and voiceprint recognition technology.
  • ASR Automatic Speech Recognition
  • TTS Text To Speech
  • Natural language processing technology including research on various theories and methods that can realize effective communication between humans and computers using natural language. Research in this area will involve natural language, that is, the language that people use every day, and usually includes technologies such as text processing, semantic understanding, machine translation, robot question answering, and knowledge graphs.
  • Machine learning is a multi-field interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. Specializes in the study of how computers simulate or implement human learning behaviors to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance.
  • Machine learning is the core of AI and the fundamental way to make computers intelligent, and its application pervades all fields of artificial intelligence.
  • Machine learning/deep learning usually includes techniques such as artificial neural network, belief network, reinforcement learning, transfer learning, inductive learning, and teaching learning.
  • related technologies of machine learning will be involved.
  • the embodiment of the present application also relates to artificial intelligence cloud service and blockchain (Blockchain).
  • the so-called artificial intelligence cloud service is generally also called AIaaS (AI as a Service, Chinese is "AI as a service”).
  • AIaaS AI as a Service
  • the AIaaS platform will split several types of common AI services and provide independent or packaged services on the cloud.
  • This service model is similar to opening an AI-themed mall: all developers can access one or more artificial intelligence services provided by the platform through the API interface, and some senior developers can also use
  • the AI framework and AI infrastructure provided by the platform are used to deploy and maintain exclusive cloud artificial intelligence services.
  • the embodiment of the present application mainly involves recommending multimedia resources to users through a multimedia recommendation platform (ie, artificial intelligence cloud service).
  • Blockchain is a new application model of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. It is essentially a decentralized database, which is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of the information (anti-counterfeiting ) and generate the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the electronic device can obtain the user's social relationship from the blockchain network, and then recommend multimedia resources to the target user based on the reliable social relationship; Resource feature information is uploaded to the blockchain for subsequent use. For example, within a period of time, it may be necessary to match the resource feature information of a certain multimedia resource with the social feature information of multiple users, then uploading the resource feature information of the multimedia resource to the blockchain can facilitate other network nodes It is directly used when recommending multimedia resources.
  • FIG. 1 is a recommended scenario diagram of a multimedia resource provided by an embodiment of the present application.
  • a multimedia resource recommendation scenario includes a terminal device 101 and a server 102 .
  • the terminal device 101 is a device used by the target user, and the terminal device 101 may include but not limited to: smart phones (such as Android phones, iOS phones, etc.), tablet computers, portable personal computers, mobile Internet devices (Mobile Internet Devices, MID ) and other equipment;
  • the terminal equipment 101 is configured with a display device, which may also be a display, a display screen, a touch screen, etc., and the touch screen may also be a touch screen, a touch panel, etc., which are not limited in this embodiment of the application.
  • the server 102 refers to a background device capable of recommending personalized multimedia resources according to the identification of the target user sent by the terminal device 101 . After determining the first multimedia resource recommended to the target user according to the target user identifier sent by the terminal device 101 , the server 102 may return the first multimedia resource to the terminal device 101 .
  • the page 103 is a schematic diagram of a page displayed by the terminal device 101 according to the first multimedia resource sent by the server 102 provided in this application.
  • the first multimedia resource is a video resource, which is displayed on the recommended subpage 1031 of the multimedia resource platform, and the recommended subpage 1031 also includes various controls related to video resources, such as follow 1032, like 1033, comment 1034, share 1035, etc.
  • the server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, and can also provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, Cloud servers for basic cloud computing services such as middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.
  • multiple servers can also be formed into a blockchain network, and each server is a node in the blockchain network.
  • the terminal device 101 and the server 102 may be directly or indirectly connected through wired communication or wireless communication, which is not limited in this application.
  • the number of terminal devices and servers in the recommendation scenario of multimedia resources shown in FIG. 1 is only an example.
  • the number of terminal devices and servers may be multiple. limited.
  • the multimedia resource recommendation scenario may also include only the terminal device 101 equipped with a multimedia resource processing device. After the user opens the multimedia resource platform, the terminal device 101 uses the multimedia resource processing device to send information to the target user. For multimedia resource recommendation, the recommended multimedia resource is displayed on page 103 .
  • the multimedia resource recommendation process mainly includes the following steps:
  • the server 102 obtains interaction data between terminal devices of multiple users; constructs a social relationship network graph according to the interaction data; and obtains an interaction vector of a target user and a set of interaction vectors of adjacent users according to the social relationship network graph.
  • the interaction vector set of adjacent users includes the interaction vector of the first adjacent user of the target user.
  • the so-called first adjacent user refers to the user who interacts with the target user (such as the target user's friends, colleagues, family members, etc.).
  • the so-called second adjacent user refers to the first adjacent user of the target user.
  • a user who has interactive behavior and does not have interactive behavior with the target user such as a friend or colleague of the target user).
  • the relationship between the target user and the first adjacent user is divided into K first relationship types, for example, the target user's friends are divided into the first first relationship type, and the target user's family is divided into K In the second first relationship type, colleagues of the target user are classified into the third first relationship type, where K is a positive integer.
  • the server 102 obtains the social feature information of the target user according to the interaction vector of the target user and the set of interaction vectors of adjacent users. The relationship feature information is determined.
  • the server 102 calculates the relationship characteristic information corresponding to each first relationship type among the K first relationship types according to the interaction vector of the target user and the interaction vector set of adjacent users; through the K first relationship
  • the K pieces of relationship feature information corresponding to the types respectively represent the social feature information of the target user, and obtain the social feature information of the target user.
  • the social feature information of the target user is jointly represented by K relationship feature information corresponding to the K first relationship types, that is, the social feature information of the target user is based on the relationship between the target user and the adjacent users of the target user.
  • the relationship type is obtained.
  • the server 102 obtains the multimedia resource set. For example, obtain the latest 50 multimedia resources released on the multimedia resource platform, the 30 multimedia resources with the most clicks in the target time period, and the 20 multimedia resources with the most accumulated clicks, etc.
  • the terminal device of the user sends the first multimedia resource in the multimedia resource set as the recommended multimedia resource.
  • the multimedia resource set whose matching degree with the target user's social feature information is higher than the matching degree threshold that is, the multimedia resource that the target user is most likely to be interested in, as the recommended first multimedia resource. resource.
  • the resource feature information of each multimedia resource in the multimedia resource set is obtained through the interaction vector of each user who has visited the multimedia resource.
  • FIG. 2 is a schematic flowchart of a method for processing multimedia resources provided by an embodiment of the present application.
  • the method described in the embodiment of this application is applied to electronic equipment, such as the terminal equipment used by some of the above-mentioned users, such as the terminal equipment 101 in Figure 1 above, or some terminal equipment with special functions server, such as the server 102 in FIG. 1 above.
  • the method includes the following steps.
  • S200 Acquire interaction data between terminal devices of multiple users, and construct a social relationship network graph according to the interaction data.
  • each user's terminal device interacts based on the interactive platform
  • various interactive behaviors are generated. For example, on social platforms, users chat, share, join groups, buy and sell items, etc. with their friends. These interactions generate corresponding interaction data, such as chat records, file sharing records, label information, transaction records, etc. Based on these interaction data, a network graph of social relationships between multiple users can be constructed.
  • the interaction vector set of adjacent users includes the interaction vector of the first adjacent user of the target user, and the interaction vector of adjacent users
  • the set includes interaction vectors of first adjacent users of the target user, and the relationship between the target user and the first adjacent users is divided into K first relationship types, where K is a positive integer.
  • the so-called first adjacent user refers to a user who interacts with the target user (such as a friend, colleague, family member, etc. of the target user), and the so-called second adjacent user refers to the first user with the target user. Users who interact with adjacent users but do not interact with the target user (such as friends and colleagues of the target user).
  • the first relationship type refers to the type of the relationship between the target user and the first adjacent user. It should be noted that the division of the K first relationship types may be set according to actual requirements. For example, the K first relationship types may be classified according to social relations, or according to the cumulative duration of interaction behaviors, or according to the time when the target user becomes friends with the first adjacent user (generated first times of interactions), and so on.
  • the friends of the target user are classified into the first relationship type, and the family members of the target user are classified into the second type.
  • the first relationship type classify the target user's colleagues in the third first relationship type.
  • S202 Obtain the social feature information of the target user according to the interaction vector of the target user and the interaction vector set of adjacent users, where the social feature information is the relationship feature corresponding to each first relationship type of the target user among the K first relationship types Information is determined.
  • the interaction vector of the target user is used to represent the characteristics of the target user in the interaction behavior.
  • the interaction vector of the target user can be obtained based on the user characteristic information carried by itself, where the user characteristic information includes the location of the target user.
  • the information of other users in various relationships can also be obtained based on the interaction vector of the first adjacent user of the target user, or based on the interaction vectors from the first adjacent user to the Sth adjacent user of the target user , S is a positive integer. It can be understood that S is proportional to the amount of user characteristic information carried in the interaction vector of the target user.
  • the interaction vector of the first adjacent user of each target user can be obtained based on the user characteristic information carried by the first adjacent user itself, or can be obtained based on the first phase
  • the interaction vector of the first adjacent user of the adjacent user may also be obtained based on the interaction vectors of the first adjacent user to the Sth adjacent user of the first adjacent user.
  • the social feature information of the target user may be a feature vector or a feature matrix.
  • the relationship feature information corresponding to each first relationship type among the K first relationship types is calculated; after obtaining the K first relationship types Among the relationship types, after the relationship characteristic information corresponding to each first relationship type, the target user is analyzed through the K relationship characteristic information respectively corresponding to the K first relationship types, and the social characteristic information of the target user is obtained. That is to say, the social feature information of the target user is jointly represented by K relationship feature information corresponding to the K first relationship types, that is, the social feature information of the target user is based on each relationship type between the target user and neighboring users owned.
  • S203 Acquire a set of multimedia resources, and display the first multimedia resource in the set of multimedia resources on the terminal device of the target user according to the social feature information of the target user.
  • the multimedia resource set may be preset, or may be obtained by the multimedia resource platform according to the real-time update of the multimedia resources in the database.
  • the resource feature information of each multimedia resource in the multimedia resource set is obtained through interaction vectors of users who have visited the multimedia resource. For example, 50 interaction vectors of users who have clicked on the multimedia resource are selected, and the resource feature information of the multimedia resource is calculated through these 50 interaction vectors.
  • the social feature information of the target user and the resource feature information of each multimedia resource in the multimedia resource set display the first multimedia resource in the multimedia resource set on the terminal device of the target user, where the first multimedia resource is a multimedia resource set One or more multimedia resources whose matching degree with the target user's social feature information is higher than the matching degree threshold, that is, the multimedia resources that the target user is most likely to be interested in.
  • the interaction data between the terminal devices of multiple users is acquired; a social relationship network diagram is constructed according to the interaction data; an interaction vector of the target user and a set of interaction vectors of adjacent users are acquired according to the social relationship network diagram, According to the interaction vector of the target user and the interaction vector set of adjacent users, the social feature information of the target user is obtained, and the social feature information is based on the K first relationship types between the target user and adjacent users, corresponding to each first relationship type If the relationship feature information is determined, the multimedia resource set is acquired, and the first multimedia resource in the multimedia resource set is sent to the target user's terminal device according to the target user's social feature information.
  • the user's social network contains a large number of friends with the same interests, such as basketball, badminton, swimming, etc., and the user usually has similar circles with friends, such as: financial circle, student circle, scientific research circle, consumption level, etc., through
  • the type of relationship between the target user and the adjacent users of the target user separates the social relationship of the user, so that the interest points of the target user can be mined from the perspective of different circles and interest groups, so that the user can be more in-depth and comprehensive to represent the interactive behavior.
  • the social characteristics of users are aggregated and multimedia resources are predicted, so that even under the condition of insufficient browsing behavior of users, social information can be better utilized and multimedia resources can be provided more accurately.
  • FIG. 3 is a schematic flowchart of another method for processing multimedia resources provided by an embodiment of the present application.
  • the method described in the embodiment of this application is applied to electronic equipment, such as the terminal equipment used by some of the above-mentioned users, such as the terminal equipment 101 in Figure 1 above, or some terminal equipment with special functions server, such as the server 102 in FIG. 1 above.
  • the method includes the following steps.
  • S301 Obtain a social relationship information set, and generate a social relationship network graph according to the social relationship information set.
  • the social relationship information collection includes user information collection and relationship information collection.
  • the user information collection includes user information of each user, for example, each user's gender, age, hobbies, and so on.
  • the electronic device generates N network nodes according to the set of user information, each of the N network nodes corresponds to a user, and each network node carries the user information of the user corresponding to the network node, and N is a positive integer;
  • the connection between each network node is determined according to the interactive behavior between users corresponding to each network node.
  • the relationship information set includes information related to the social relationship formed by two users, for example, interaction behavior records, group information, label information, and the like. If the relationship information set indicates that the user corresponding to the first network node among the N network nodes and the user of the second network node have interactive behaviors (such as chatting, transfer, social circle interaction, etc.), then according to the interactive behavior, generate the first network node and the edge information of the second network node to obtain a social relationship network graph.
  • interactive behaviors such as chatting, transfer, social circle interaction, etc.
  • the edge information of the first network node and the second network node includes edge weights.
  • the electronic device can determine the edge weights of each edge in the social relationship network graph according to the relationship information set. Specifically, according to the degree of association between the first network node and the second network node, the weight of the connection is determined, and the weight of the connection is proportional to the degree of association. It is determined by the interaction information in , and the interaction information includes: at least one item of cumulative interaction times, cumulative interaction duration, interaction frequency, and interaction content.
  • the weight of the link between user A and user B is greater than The weight of the connection between user A and user C.
  • first define the user social relationship network graph G (A, E).
  • the number of all users is N
  • A is the user association matrix
  • E is the user characteristic information.
  • the electronic device connects the network nodes corresponding to each user into a user social relationship network graph.
  • the edge weight between users is determined by the degree of user association.
  • the measurement of user interaction behavior can be jointly determined by variables such as the number of user interaction behaviors, the cumulative duration of interaction behaviors, interaction discussion, interaction frequency ranking, and the number of interaction days in the last week.
  • Fig. 4a is a schematic diagram of a user social relationship network diagram provided by an embodiment of the present application. As shown in Figure 4a, assuming that the users corresponding to network nodes u1 and u2 have more interactions, and the users corresponding to network nodes u1 and u3 have fewer interactions, the weight of the connection between network nodes u1 and u2 is higher than that of network nodes The edge weight between u1 and u3 is high. Assuming that the user corresponding to network node u1 has no interaction with users corresponding to other network nodes except network nodes u2 and u3, then u1 has no connection with other network nodes.
  • the number of interactions between two users can be recorded as c ij
  • S302 Obtain an interaction vector of the target user and a set of interaction vectors of adjacent users according to the social relationship network graph.
  • step S302 After obtaining the social relationship network graph, in this step S302, based on the social relationship network graph, the user's social relationship is separated to obtain the interaction vector of the target user and a set of interaction vectors of each adjacent user.
  • the electronic device obtains the interaction vectors of N users corresponding to N network nodes according to the social relationship network graph.
  • the purpose is to describe the interaction behavior of users in the social relationship network through vectors, so that the vectors of users with close social relationships
  • the representations are relatively similar; correspondingly, the vector representations of users with social distance are quite different.
  • the interaction vector of the target user is obtained.
  • the interaction vector x i of the target user can be expressed as:
  • the probability of walking from the i-th network node to the j-th network node is proportional to the edge weight in the edge information of the i-th network node and the j-th network node, i and j are both positive integers , i is not equal to j, and both i and j are less than or equal to N. That is to say, the larger A ij is, the higher the probability of walking from network node i to network node j is.
  • the electronic device represents the interaction vector for the user through methods such as vectorized embedding.
  • vectorized embedding methods include unsupervised user embedding methods such as Node2Vec node embedding.
  • Node2Vec node embedding method As an example, starting from the target network node in the social network graph, multiple trajectories are walked; then, all the trajectories walked out are used as a corpus and input into the word2vec word vector embedding algorithm model. Through the word2vec word vector embedding algorithm model, the corpus is processed to obtain the interaction vector of the target user corresponding to the target network node.
  • the edge weights between network nodes corresponding to different users are different in the social relationship network graph, in the process of vectorization embedding, the influence of weights can be considered, and the walking method with weights can be used. In this way, the probability of walking from network node i to network node j is proportional to A ij .
  • the electronic device can obtain a matrix composed of interaction vectors of users corresponding to all nodes in the social network graph, denoted as X,
  • xi represents the interaction vector of user i.
  • the electronic device can extract the interaction vector of the target user and the interaction vector of adjacent users from the matrix X gather.
  • Fig. 4b is a schematic diagram of a graph convolution model based on a social relationship network graph provided by an embodiment of the present application.
  • V0 is the target network node corresponding to the target user
  • V1-V8 are network nodes corresponding to the first neighboring users of the target user.
  • V1-V8 have the same aggregation weight and the same mapping function. That is to say, regardless of the degree of correlation between V1-V8 and V0, it is considered that V1-V8 has the same influence on V0. For example, if a certain group or relationship group is an object rarely browsed by target users, and each adjacent user in it is weakly connected to the target user, then at this time, the weight of the connection between each adjacent user and the target user is considered to be same.
  • V1-V8 are users of V0 who have established social relationships in different ways, for example, the user corresponding to network node V1 is the family member of the target user (V0), and the user corresponding to network node V5 is the target user ( A friend of V0), the network node V8 is a client of the target user (V0). Therefore, each adjacent user has different influence on the target user, and the closer the relationship is, the higher the influence on the target user is, that is, the higher the degree of association.
  • different users need to be classified, for example, users are divided into multiple categories according to factors such as the reason for establishing a connection, the type of relationship, the cumulative duration of establishing a social relationship, and the frequency of interaction.
  • this embodiment of the present application implements step S303 through a method similar to clustering.
  • the electronic device divides the first adjacent users of the target user into K first relationship types according to preset rules.
  • K first relationship types such as family members, friends, and customers are preset
  • K A feature parameter set of the h-th first relationship type in the first relationship type the feature parameter set includes the weight matrix of the h-th first relationship type and the offset vector of the h-th first relationship type.
  • the weight matrix W h of the h-th first relationship type and the bias vector b h of the h-th first relationship type are initialized. For example, random initialization is performed on the weight matrix W h of the h-th first relationship type and the bias vector b h of the h-th first relationship type. Then, during the training process, the weight matrix W h of the h-th first relationship type and the bias vector b h of the h-th first relationship type are updated through the gradient descent method, and finally the h-th first relationship is obtained Type updated weight matrix W h and bias vector b h of the hth first relation type.
  • the electronic device can obtain the weight matrix and bias vector of each first relationship type based on the above method.
  • the target user calculates intermediate features (i.e., the hidden representation of the target user) under the hth first relationship type, and calculate the target user’s first Neighboring user intermediate features (ie implicit representation of the first neighboring user) under the h-th first relation type of neighboring users.
  • each of the intermediate features of the target user and the intermediate features of adjacent users can be calculated according to the user's interaction vector x i , the weight matrix W h and the bias vector b h .
  • the implicit representation z i,h of user i in the hth first relationship type can be expressed as:
  • 2 means to calculate the modulus length of x, and the operation of dividing by the modulus length is to get rid of the influence of vector length on classification;
  • ⁇ (x) is an activation function (such as sigmoid function, tanh function, Relu function, etc.)
  • W h of the h-th first relationship type W h of the h-th first relationship type
  • x i is the interaction vector of user i (obtained from the matrix X in step S302)
  • b h is the weight matrix of the h-th first relationship type bias vector.
  • the electronic device can calculate and obtain implicit representations of the target user and each first neighboring user of the target user.
  • the relationship feature information of the hth first relationship type is obtained according to the intermediate features of the target user and the intermediate features of each neighboring user.
  • the relationship feature information c h of the first relationship type can be specifically expressed as:
  • the network node u corresponds to V0 in Figure 4b (that is, the network node corresponding to the target user), and (v
  • p v,h is used to represent the probability that user v is assigned to the h-th first relationship type
  • z u,h is the hidden representation of the target user in the h-th first relationship type
  • z v,h is the hidden representation of the target user's first neighbor user in the h-th first relationship type.
  • the value of p v,h is determined according to the quantity of the first relationship type. In a specific embodiment, if the number of the first relationship type of user A is 5 (that is, the social relationship of user A is divided into 5 categories), then
  • the electronic device can calculate and obtain K relationship feature information of the first relationship type.
  • the relationship feature information of the hth first relationship type is obtained according to the intermediate features of the target user and the intermediate features of each adjacent user.
  • the relationship feature information of the h-th first relationship type is the relationship feature information obtained at the T-th iteration of the h-th first relationship type, where T is a positive integer.
  • the electronic device uses an iterative method to obtain the target probability that each first adjacent user of the target user is divided into the h-th first relationship type at the t-th iteration t is a positive integer, and t is less than T; then the electronic device according to the target probability and the intermediate features of each adjacent user of the target user (that is, the implicit representation of the first adjacent user (z v,h )), calculate the aggregated feature of the first adjacent user of the target user Perform calculation processing on the intermediate features of the target user (that is, the implicit representation of the target user (z u,h )) and the aggregated features of the target user’s first adjacent users to obtain the h-th first relationship type at the t+1 iteration
  • the relationship feature information obtained at the time is the target probability that each first adjacent user of the target user is divided into the h
  • the target probability It can be expressed as:
  • the exponential function exp(x) means to calculate the index of x
  • the network node u corresponds to V0 in Figure 4b (that is, the network node corresponding to the target user), and (v
  • the initial value of the probability that user v is assigned to the hth first relationship type It is determined according to the quantity of the first relationship type.
  • the electronic device can calculate the relationship feature information obtained at the T-th iteration of each first relationship type, where the T-th iteration is the last calculation, so The relationship feature information of the K first relationship types is obtained.
  • S304 Perform feature analysis on the target user through K relationship feature information corresponding to the K first relationship types to obtain social feature information of the target user.
  • the social feature information is a set of K pieces of relationship feature information.
  • the electronic device acquires the first feature information set of the first adjacent user of the target user, and the second feature information set of the second adjacent user of the target user, wherein the second adjacent user refers to
  • the first feature information set includes each of the multiple second relationship types of the first adjacent user of the target user, each second relationship type Corresponding relationship feature information (such as [c′ 1 ,c′ 2 ,...,c′ R ])
  • the second feature information set includes multiple third relationship types of the second adjacent users of the target user, each of the third relationship types
  • the relationship feature information corresponding to the three relationship types (such as [c′′ 1 ,c′′ 2 ,...,c′′ S ]);
  • R, S are positive integers, and R, S, K can be the same or different.
  • step S301-step S303 Let me repeat.
  • the social feature information of the first adjacent user of the target user can be obtained Similarly, through the second feature information set, the social feature information of the second adjacent user of the target user can be obtained.
  • the social feature information of the L-th adjacent user of the target user can be obtained
  • the relationship characteristic information corresponding to the K first relationship types of the target user, the first characteristic information set to the Lth characteristic information set are input into the relationship prediction model, and the prediction result output by the relationship prediction model is obtained.
  • L 2.
  • the prediction result is over-fitted to obtain the social feature information of the target user.
  • the relationship prediction model includes an L-layer graph convolutional network layer, and L is a positive integer.
  • the graph convolutional model shown in Figure 4b above represents a layer of graph convolutional network layers, and the relationship prediction model includes a total of L cascaded graph convolutional network layers, where the input data of the gth graph convolutional network layer includes : The data obtained after overfitting the output data of the g-1th layer graph convolutional network layer.
  • the input of layer l of the relationship prediction model is the processing result of the target user at layer l-1
  • the processing result set of the first adjacent user of the target user at layer l-1 l is a positive integer, and l is less than or equal to L
  • the input data can be processed by using the lth layer of the relational prediction model as follows:
  • over-fitting is performed on the output data of the first layer of the relationship prediction model to obtain the processing result of the target user at the first layer
  • f (l) (x) means to process x through the l layer of the relationship prediction model
  • dropout (x) means to overfit x
  • the value of is initialized to x u .
  • the processing results of target users at layer L It can be expressed as:
  • U u covers more feature information than y u .
  • S305 Acquire a set of multimedia resources, and display the first multimedia resource in the set of multimedia resources on the terminal device of the target user according to the social feature information of the target user.
  • the multimedia resource set includes resource characteristic information of the second multimedia resource.
  • the electronic device obtains the access record of the second multimedia resource, the access record includes the user identifiers of Q users who have accessed the second multimedia resource, and Q is a positive integer; according to the user identifiers of the Q users, obtain the Q user IDs interaction vector (such as obtained from the matrix X in step S302); the interaction vectors of Q users are fused, such as superimposing the interaction vectors of Q users, and performing mean pooling (mean pooling) processing to obtain resource characteristic information of the second multimedia resource, wherein the resource characteristic information of the second multimedia resource is included in the multimedia resource set.
  • mean pooling mean pooling
  • the list of users accessing the second multimedia resource is (u 1 , u 2 , u 3 , u 4 ,...u H ), where H is the total number of all users who have accessed the video. If the resource characteristic information of the second multimedia resource is expressed as im , then:
  • the electronic device calculates the matching degree between the target user and each multimedia resource in the multimedia resource set according to the social feature information of the target user and the resource feature information of each multimedia resource in the multimedia resource set, wherein the first multimedia resource is The multimedia resource with the highest degree of matching with the target user in the multimedia resource collection.
  • the electronic device can predict the relationship between each first relationship type and the multimedia resource i through a multi-layer perceptron (Multi-Layer Percetron, MLP), and then use the attention mechanism to comprehensively consider different first relationship types Genre preferences, such as the target user's colleagues like to watch costume dramas, and the target user's family members like to watch urban dramas, and finally get the prediction results of target user u and multimedia resource i.
  • MLP Multi-Layer Percetron
  • the electronic device splices the resource feature information of the second multimedia resource in the multimedia resource set and the social feature information of the target user to obtain a spliced feature set, that is, combines the resource feature information of the second multimedia resource with the target user's social feature information.
  • the relationship feature information corresponding to each first relationship type in [c u,1 , cu,2 ,..., cu,K ] is spliced to obtain K spliced features.
  • a multi-layer perceptron is used to process each concatenated feature in the concatenated feature set to obtain a relationship vector between each of the first relationship types and the second multimedia resource among the K first relationship types.
  • the relationship vector between the kth first relationship type of the target user and the multimedia resource i can be expressed as:
  • y represents the concatenation of vector x and vector y
  • MLP 1 (x) represents the processing of x by the first multi-layer perceptron.
  • the smart device can process each splicing feature in the splicing feature set through the first multi-layer perceptron to obtain the relationship vector between each first relationship type and each multimedia resource.
  • the electronic device After obtaining the relationship vectors between each first relationship type and the second multimedia resource, the electronic device calculates the weight corresponding to the first relationship type according to the relationship vector between each first relationship type and the second multimedia resource.
  • the weight of the kth first relationship type of the target user and the multimedia resource i can be expressed as:
  • the exponential function exp(x) means to calculate the exponent of x
  • ⁇ (x) is the activation function (such as the sigmoid function)
  • a T is the attention vector
  • a T r u,i,k means the relationship vector and the attention vector Perform dot multiplication.
  • the electronic device may calculate weights between each first relationship type of the target user and each multimedia resource.
  • the electronic device After obtaining the weight between each first relationship type of the target user and the second multimedia resource, the electronic device according to the relationship vector between each first relationship type and the second multimedia resource among the K first relationship types , and the weight corresponding to each first relationship type, to obtain the matching degree between the target user and the second multimedia resource.
  • the matching degree between the target user and the multimedia resource i can be expressed as:
  • MLP 2 (x) means that the second multi-layer perceptron is used to process x. Based on the above formula (13), the electronic device can obtain the matching degree between the target user and each multimedia resource in the multimedia resource set through the second multi-layer perceptron.
  • the electronic device sorts the multimedia resources in the multimedia resource set according to the order of matching degree from high to low, and determines the multimedia resource (one or more) arranged before the target position as the first multimedia resource . Then recommend the first multimedia resource to the target user.
  • the electronic device is a server
  • the server sends the first multimedia resource to the terminal device of the target user for display; when the electronic device is a terminal device, the determined first multimedia resource is displayed on the interface, as shown in the figure 1 on page 103.
  • the electronic device before recommending multimedia resources, can use the training data to optimize the parameters in formula (3) - formula (13), that is, combine the labeled data with formula (3) -
  • the prediction data calculated by formula (13) are compared, and the parameters in formula (3) - formula (13) are adjusted through the loss function to reduce the difference between the labeled data and the predicted data until the loss function converges.
  • the electronic device After the training is completed, the electronic device acquires interaction vectors of all users.
  • the multimedia resource acquisition request of the target user u is detected, the above step S301-step S305 is executed to compare the similarity between the target user u and each multimedia resource in the multimedia resource set, and then recommend multimedia resources that meet the recommendation conditions to the target user u .
  • the embodiment of the present application obtains the relationship feature information corresponding to each first relationship type in the first relationship type through the implicit representation of the target user and the first adjacent user of the target user, and then obtains The social feature information of the target user; the resource feature information of the multimedia resource is obtained through the interaction vector of the user who has watched the multimedia resource, and then the relationship between the social feature information of the target user and the resource feature information of the multimedia resource is determined through the multi-layer perceptron MLP matching degree, so as to recommend multimedia resources to target users.
  • the relationship between each group vector and video i can be predicted, and then through the attention mechanism, the interests of different group vectors are comprehensively considered, and finally the prediction results of user u and video i are obtained.
  • This prediction method based on the matching degree between users and videos based on the attention mechanism can better model user u's interest in video i from the perspective of different social relationships, so that the recommendation of multimedia resources can be separated from historical browsing data. , while ensuring the accuracy of the recommendation.
  • FIG. 5 is a schematic structural diagram of a multimedia resource processing device provided by an embodiment of the present application.
  • the multimedia resource processing device 500 includes an acquisition unit 501 and a processing unit 502.
  • the device can be mounted on an electronic device.
  • Devices may include terminal devices and servers.
  • the apparatus for processing multimedia resources shown in FIG. 5 may be used to execute part or all of the functions in the method embodiments described in FIG. 2 and FIG. 3 above. Among them, the detailed description of each unit is as follows:
  • the acquisition unit 501 is configured to acquire interaction data between terminal devices of multiple users; construct a social relationship network graph according to the interaction data; and acquire an interaction vector of a target user and a neighbor user's interaction vector according to the social relationship network graph.
  • a set of interaction vectors, the set of interaction vectors of adjacent users includes the interaction vectors of the first adjacent user of the target user, and the relationship between the target user and the first adjacent user is divided into K
  • the first relationship type, K is a positive integer;
  • the processing unit 502 is configured to obtain social feature information of the target user according to the interaction vector of the target user and the set of interaction vectors of the adjacent users, the social feature information is based on the target user in the K Among the first relationship types, the relationship feature information corresponding to each first relationship type is determined; acquire a multimedia resource set, and display the target user’s terminal device according to the target user’s social feature information The first multimedia resource in the collection of multimedia resources.
  • processing unit 502 is specifically configured to:
  • the interaction vector of the target user and the interaction vector set of the adjacent users calculate and obtain the relationship characteristic information corresponding to each first relationship type among the K first relationship types;
  • the social feature information is obtained by performing feature analysis on the target user through the K relationship feature information corresponding to the K first relationship types.
  • processing unit 502 is specifically configured to:
  • relationship feature information of the hth first relationship type is obtained.
  • the relationship feature information of the h-th first relationship type is the relationship feature information obtained at the T-th iteration of the h-th first relationship type, where T is a positive integer; the processing unit 502, specifically for:
  • processing unit 502 is specifically configured to:
  • the first feature information set includes: among the multiple second relationship types of the first adjacent user, each second relationship type corresponds to The relationship feature information of the second feature information set includes: among the multiple third relationship types of the second adjacent user, the relationship feature information corresponding to each third relationship type;
  • relationship feature information corresponding to the K first relationship types of the target user input the relationship feature information corresponding to the K first relationship types of the target user, the first set of feature information, and the second set of feature information into a relationship prediction model to obtain a prediction output by the relationship prediction model result;
  • the relationship prediction model includes an L-layer graph convolutional network layer, L is a positive integer, and the input data of the g-th layer graph convolutional network layer includes: for the g-1th layer graph convolutional network layer The data obtained after the output data is over-fitted.
  • processing unit 502 is specifically configured to:
  • the access record includes user identifiers of Q users who have accessed the second multimedia resource, where Q is a positive integer;
  • processing unit 502 is specifically configured to:
  • the social feature information of the target user and the resource feature information of each multimedia resource in the multimedia resource set calculate the matching degree between the target user and each multimedia resource in the multimedia resource set, wherein the first The multimedia resource is the multimedia resource with the highest matching degree with the target user in the multimedia resource set.
  • processing unit 502 is specifically configured to:
  • the relationship vector between each first relationship type and the second multimedia resource, and the weight corresponding to each first relationship type, the relationship between the target user and the second multimedia resource is obtained. 2. Matching degree of multimedia resources.
  • the acquiring unit 501 is specifically configured to:
  • the set of social relationship information includes a set of user information and a set of relationship information
  • each of the N network nodes corresponds to a user, and each network node carries the user information of the user corresponding to the network node, and N is positive integer;
  • the relationship information set indicates that the user corresponding to the first network node among the N network nodes has interactive behavior with the user of the second network node, then according to the interactive behavior, generate the first network node and the The edge information of the second network node is used to obtain the social relationship network graph.
  • the edge information includes edge weights of the first network node and the second network node, and the edge weights are based on the weights of the first network node and the second network node Determined by the degree of association between them, the edge weight is proportional to the degree of association.
  • the association degree is determined according to the interaction information between the first network node and the second network node within a target time period, the interaction information includes: cumulative interaction times, cumulative interaction duration, At least one of interaction frequency and interaction content.
  • the acquiring unit 501 is specifically configured to:
  • the step size of each track is P; wherein, M and P are both positive integers;
  • an interaction vector of the target user is obtained.
  • the probability of walking from the i-th network node to the j-th network node is proportional to the edge weights of the i-th network node and the j-th network node, i and j are both is a positive integer, i is not equal to j, and both i and j are less than or equal to N.
  • step S200 and step S201 shown in FIG. 2 may be executed by the acquiring unit 501 shown in FIG. 5
  • step S202 and step S203 may be executed by the processing unit 502 shown in FIG. 5
  • step S301 and step S302 shown in FIG. 3 may be executed by the obtaining unit 501 shown in FIG. 5
  • steps S303 to S305 may be executed by the processing unit 502 shown in FIG. 5 .
  • Each unit in the multimedia resource processing device shown in FIG. 5 can be separately or all combined into one or several other units to form, or one (some) units can be split into functionally smaller units. Multiple units can realize the same operation without affecting the realization of the technical effects of the embodiments of the present application.
  • the above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit may also be realized by multiple units, or the functions of multiple units may be realized by one unit. In other embodiments of the present application, the apparatus for processing multimedia resources may also include other units. In practical applications, these functions may also be implemented with the assistance of other units, and may be implemented cooperatively by multiple units.
  • a general-purpose computing device such as a computer that includes processing elements such as a central processing unit (CPU), a random access storage medium (RAM), and a read-only storage medium (ROM) and storage elements
  • CPU central processing unit
  • RAM random access storage medium
  • ROM read-only storage medium
  • the problem-solving principle and beneficial effect of the multimedia resource processing device provided in the embodiment of the present application are similar to the problem-solving principle and beneficial effect of the multimedia resource processing device in the method embodiment of the present application. Please refer to the implementation of the method The principle and beneficial effects of the invention are described briefly, so they are not repeated here.
  • FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • the electronic device 600 includes at least a processor 601 , a communication interface 602 and a memory 603 .
  • the processor 601, the communication interface 602 and the memory 603 may be connected through a bus or in other ways.
  • the processor 601 or central processing unit (Central Processing Unit, CPU) is the calculation core and control core of the terminal, which can analyze various instructions in the terminal and process various data of the terminal.
  • CPU Central Processing Unit
  • the CPU can use It is used to analyze the power-on/off instructions sent by the user to the terminal, and control the terminal to perform power-on/off operations; another example: the CPU can transmit various interactive data between the internal structures of the terminal, and so on.
  • the communication interface 602 can optionally include standard wired interfaces and wireless interfaces (such as WI-FI, mobile communication interfaces, etc.), which can be used to send and receive data under the control of the processor 601; the communication interface 602 can also be used for internal data transfer of the terminal transmission and interaction.
  • the memory 603 (Memory) is a storage device in the terminal, and is used to store programs and data.
  • the memory 603 here may include not only a built-in memory of the terminal, but also an extended memory supported by the terminal.
  • the memory 603 provides a storage space, which stores the operating system of the terminal, which may include but not limited to: Android system, iOS system, Windows Phone system, etc., which is not limited in this application.
  • the processor 601 executes the following operations by running the executable program code in the memory 603:
  • an interaction vector of the target user and a set of interaction vectors of adjacent users are obtained, the interaction vector set of adjacent users includes an interaction vector of a first adjacent user of the target user, the The relationship between the target user and the first adjacent user is divided into K first relationship types, where K is a positive integer;
  • the social feature information of the target user is obtained, and the social feature information is based on the target user in the K first relationship types , determined by the relationship characteristic information corresponding to each first relationship type;
  • the problem-solving principle and beneficial effect of the electronic device provided in the embodiment of the present application are similar to the problem-solving principle and beneficial effect of the multimedia resource processing method in the method embodiment of the present application. Please refer to the principle and beneficial effect of the implementation of the method The beneficial effects are described briefly and will not be repeated here.
  • the embodiment of the present application also provides a computer-readable storage medium, wherein one or more instructions are stored in the computer-readable storage medium, and the one or more instructions are suitable for being loaded by a processor and executing the method described in the above-mentioned embodiment.
  • the embodiment of the present application also provides a computer program product containing instructions, which, when run on a computer, causes the computer to execute the method for processing multimedia resources described in the above method embodiments.
  • the embodiment of the present application also provides a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium.
  • the processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above multimedia resource processing method.
  • the modules in the device of the embodiment of the present application can be combined, divided and deleted according to actual needs.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Library & Information Science (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请实施例公开了一种多媒体资源的处理方法、装置、设备及存储介质。其中方法包括：获取多个用户的终端设备之间的交互数据；根据所述交互数据，构建社交关系网络图；根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；及，获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。

Description

多媒体资源的处理方法、装置、设备及存储介质
本申请要求于2021年6月17日提交中国专利局、申请号为202110675393.6、申请名称为“一种对多媒体资源的推荐方法、装置、设备及存储介质”的中国专利申请的优先权。
技术领域
本申请涉及计算机技术领域，具体涉及一种多媒体资源的处理方法、装置、设备及计算机可读存储介质。
发明背景
随着计算机技术的发展，网络中涌现出来海量的多媒体资源。目前，大多数多媒体资源的推荐算法(如协同过滤算法)是基于用户的历史浏览记录，来挖掘(发现)用户的兴趣点，进而向用户推荐其可能感兴趣的多媒体资源。
实践发现，现有的推荐算法对用户的历史浏览记录的依赖性较强。当用户实际浏览行为较少时，如当前用户为新用户，或者平台的业务处于起步阶段，能够获取到的该用户的浏览记录非常有限，使得基于现有的推荐算法推荐的多媒体资源，与当前用户实际感兴趣的多媒体资源的匹配度较低。因此，在这种场景下，现有的推荐算法的推荐准确度较低。
发明内容
本申请实施例提供了一种多媒体资源的处理方法、装置、设备及存储介质，能够在缺乏历史浏览数据的情况下，有效地提高推荐准确度。
一方面，本申请实施例提供了一种多媒体资源的处理方法，由电子设备执行，包括：
获取多个用户的终端设备之间的交互数据；
根据所述交互数据，构建社交关系网络图；
根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；及，
获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
一方面，本申请实施例还提供了一种多媒体资源的处理装置，包括：
获取单元，用于获取多个用户的终端设备之间的交互数据；根据所述交互数据，构建社交关系网络图；根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
处理单元，用于根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
相应地，本申请实施例还提供了一种电子设备，包括：存储装置和处理器；所述存储装置中存储有计算机程序；处理器，执行计算机程序，实现上述的多媒体资源的处理方法。
相应地，本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计 算机程序，所述计算机程序被处理器执行时，实现上述的多媒体资源的处理方法。
相应地，本申请提供了一种计算机程序产品或计算机程序，该计算机程序产品或计算机程序包括计算机指令，该计算机指令存储在计算机可读存储介质中，计算机设备的处理器从计算机可读存储介质读取该计算机指令，处理器执行该计算机指令，使得该计算机设备执行上述的多媒体资源的处理方法。
附图简要说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种多媒体资源的处理场景图；
图2为本申请实施例提供的一种多媒体资源的处理方法的流程示意图；
图3为本申请实施例提供的另一种多媒体资源的处理方法的流程示意图；
图4a为本申请实施例提供的一种用户社交关系网络图的示意图；
图4b为本申请实施例提供的一种基于社交关系网络图的图卷积模型的示意图；
图5为本申请实施例提供的一种多媒体资源的处理装置的结构示意图；
图6为本申请实施例提供的一种电子设备的结构示意图。
实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述。
本申请实施例涉及人工智能(Artificial Intelligence，AI)及机器学习(Machine Learning，ML)。其中，AI是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说，人工智能是计算机科学的一个综合技术；其主要通过了解智能的实质，生产出一种新的能以人类智能相似的方式做出反应的电子设备，使得电子设备具有感知、推理与决策等多种功能。
本申请实施例提供的电子设备，能够基于目标用户与目标用户的相邻用户的关系类型，来向浏览行为较少的目标用户进行多媒体资源推荐。
AI技术是一门综合学科，涉及领域广泛，既有硬件层面的技术也有软件层面的技术。AI基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大应用程序的处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、语音处理技术、自然语言处理(Nature Language processing,NLP)技术以及机器学习/深度学习等几大方向。
本申请实施例在构建社交关系网络图的过程中，在将用户间的交互行为转换为连边信息时，会涉及上述一种或多种AI软件技术。例如，在提取用户间发送的(短)视频的特征时，会涉及计算机视觉技术；在提取用户间发送的语音的特征时，会涉及语音处理技术；在提取用户间发送的文本信息的特征时，会涉及自然语言处理技术。
其中，计算机视觉技术通常包括图像处理、图像识别、图像语义理解、图像检索、光学字符识别(Optical Character Recognition，OCR)、视频处理、视频语义理解、视频内容/行为识别、三维物体重建、3D技术、虚拟现实、增强现实、同步定位与地图构建等技术，还包括常见的人脸识别、指纹识别等生物特征识别技术。
语音处理技术包括自动语音识别技术(Automatic Speech Recognition，ASR)和语音合成技术(Text To Speech，TTS)以及声纹识别技术。
自然语言处理技术，包括研究能实现人与计算机之间用自然语言进行有效通信的各种理论和方法。这一领域的研究将涉及自然语言，即人们日常使用的语言，通常包括文本处理、语义理解、机器翻译、机器人问答、知识图谱等技术。
机器学习是一门多领域交叉学科，涉及概率论、统计学、逼近论、凸分析、算法复杂度理论等多门学科。专门研究计算机怎样模拟或实现人类的学习行为，以获取新的知识或技能，重新组织已有的知识结构使之不断改善自身的性能。机器学习是AI的核心，是使计算机具有智能的根本途径，其应用遍及人工智能的各个领域。机器学习/深度学习通常包括人工神经网络、置信网络、强化学习、迁移学习、归纳学习、式教学习等技术。本申请实施例在根据社交关系网络图，训练关系预测模型的过程中，会涉及机器学习的相关技术。
此外，本申请实施例还涉及人工智能云服务和区块链(Blockchain)。所谓人工智能云服务，一般也被称作是AIaaS(AI as a Service，中文为“AI即服务”)。具体来说，AIaaS平台会把几类常见的AI服务进行拆分，并在云端提供独立或者打包的服务。这种服务模式类似于开了一个AI主题商城：所有的开发者都可以通过API接口的方式，来接入使用平台提供的一种或者是多种人工智能服务，部分资深的开发者还可以使用平台提供的AI框架和AI基础设施，来部署和运维专属的云人工智能服务。本申请实施例主要涉及通过多媒体推荐平台(即人工智能云服务)，向用户进行多媒体资源推荐。
区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。其本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层。
在本申请实施例中，电子设备可以从区块链网络中获取用户的社交关系，进而基于可靠的社交关系向目标用户推荐多媒体资源；也可以将分析得到的用户的社交特征信息和多媒体资源的资源特征信息上传至区块链，以便于后续使用。例如，在一个时间段内，可能需要某个多媒体资源的资源特征信息与多个用户的社交特征信息进行匹配，那么，将多媒体资源的资源特征信息上传至区块链，可以方便协助其他网络节点在进行多媒体资源推荐时直接使用。
请参见图1，图1为本申请实施例提供的一种多媒体资源的推荐场景图。如图1所示，多媒体资源的推荐场景中包括了终端设备101和服务器102。其中，终端设备101为目标用户所使用的设备，终端设备101可以包括但不限于：智能手机(如Android手机、iOS手机等)、平板电脑、便携式个人计算机、移动互联网设备(Mobile Internet Devices，MID)等设备；终端设备101配置有显示装置，显示装置也可为显示器、显示屏、触摸屏等等，触摸屏也可为触控屏、触控面板等等，本申请实施例不做限定。
服务器102是指能够根据终端设备101发送的目标用户的标识推荐个性化多媒体资源的后台设备。在根据终端设备101发送的目标用户的标识确定向目标用户推荐的第一多媒体资源后，服务器102可以向终端设备101返回该第一多媒体资源。
页面103为本申请提供的一种终端设备101根据服务器102发送的第一多媒体资源进行显示的页面示意图。其中，所述第一多媒体资源是一种视频资源，显示在多媒体资源平台的推荐子页面1031中，该推荐子页面1031还包括视频资源相关的各种控件，例如，关注1032、点赞1033、评论1034、分享1035等。
服务器102可以是独立的物理服务器，也可以是多个物理服务器构成的服务器集群或者分布式系统，还可以是提供云服务、云数据库、云计算、云函数、云存储、网络服务、云通信、中间件服务、域名服务、安全服务、CDN、以及大数据和人工智能平台等基础云计算服务的云服务器。此外，还可以将多个服务器组成为一个区块链网络，每个服务器为区块链网络中的一个节点。
终端设备101和服务器102之间可以通过有线通信或者无线通信方式进行直接或间接地连接，本申请在此不做限制。
需要说明的是，图1所示的多媒体资源的推荐场景中终端设备和服务器的数量仅为举例，例如，终端设备和服务器的数量可以为多个，本申请并不对终端设备和服务器的数量进行限定。
可选的，多媒体资源的推荐场景中也可以只包括搭载有多媒体资源的处理装置的终端设备101，终端设备101在用户打开多媒体资源平台后，通过搭载的多媒体资源的处理装置，向目标用户进行 多媒体资源推荐，如在页面103中显示所推荐的多媒体资源。
图1所示的多媒体资源的推荐场景中，多媒体资源的推荐流程主要包括以下步骤：
(1)服务器102获取多个用户的终端设备之间的交互数据；根据交互数据，构建社交关系网络图；根据社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合。其中，相邻用户的交互向量集合中包括目标用户的第一相邻用户的交互向量。
所谓第一相邻用户，是指与目标用户存在交互行为的用户(如目标用户的朋友、同事、家人等)，同理，所谓第二相邻用户是指与目标用户的第一相邻用户存在交互行为，且与目标用户不存在交互行为的用户(如目标用户的朋友的同事)。
进一步地，目标用户与第一相邻用户之间的关系被划分为K个第一关系类型，例如，目标用户的朋友被划分在第一个第一关系类型中，目标用户的家人被划分在第二个第一关系类型中，目标用户的同事被划分在第三个第一关系类型中，其中，K为正整数。
(2)服务器102根据目标用户的交互向量和相邻用户的交互向量集合，得到目标用户的社交特征信息，社交特征信息是根据目标用户在K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的。
在一个实施例中，服务器102根据目标用户的交互向量和相邻用户的交互向量集合，计算K个第一关系类型中，每个第一关系类型对应的关系特征信息；通过K个第一关系类型分别对应的K个关系特征信息，对目标用户的社交特征进行表示，得到目标用户的社交特征信息。
也就是说，目标用户的社交特征信息是由K个第一关系类型分别对应的K个关系特征信息共同表示的，即，目标用户的社交特征信息是基于目标用户与目标用户的相邻用户的关系类型得到的。
(3)服务器102获取多媒体资源集合。例如，获取多媒体资源平台上最新发布的50个多媒体资源、在目标时间段点击次数最多的30个多媒体资源、累积点击次数多的20个多媒体资源等，并根据目标用户的社交特征信息，向目标用户的终端设备发送多媒体资源集合中的第一多媒体资源，作为所推荐的多媒体资源。
具体而言，确定多媒体资源集合中与目标用户的社交特征信息匹配度高于匹配度阈值的一个或多个多媒体资源，即目标用户最可能感兴趣的多媒体资源，作为推荐的第一多媒体资源。
在一个实施例中，多媒体资源集合中的每个多媒体资源的资源特征信息，是通过访问过该多媒体资源的每个用户的交互向量得到的。
请参阅图2，图2为本申请实施例提供的一种多媒体资源的处理方法的流程示意图。本申请实施例的所述方法应用于电子设备，该电子设备例如可以是上述提及的某些用户所使用的终端设备，如上述图1中的终端设备101，也可以是某些具有特殊功能的服务器，如上述图1中的服务器102。所述方法包括如下步骤。
S200：获取多个用户的终端设备之间的交互数据，根据交互数据，构建社交关系网络图。
各个用户的终端设备，基于交互平台进行交互时，产生各种交互行为。例如，在社交平台上，用户与好友之间聊天、分享、加入群组、买卖物品等。这些交互行为产生了相应的交互数据，如聊天记录、文件分享记录、标签信息、交易记录等。基于这些交互数据，可以构建多个用户之间的社交关系网络图。
S201：根据社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，相邻用户的交互向量集合中包括目标用户的第一相邻用户的交互向量，相邻用户的交互向量集合中包括目标用户的第一相邻用户的交互向量，目标用户与第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数。
本申请实施例中，所谓第一相邻用户，是指与目标用户存在交互行为的用户(如目标用户的朋友、同事、家人等)，所谓第二相邻用户是指与目标用户的第一相邻用户存在交互行为、且与目标用户不存在交互行为的用户(如目标用户的朋友的同事)。
第一关系类型是指目标用户与第一相邻用户之间的关系所属的类型。需要说明的是，K个第一关系类型的划分，可以根据实际需求进行设定。例如，K个第一关系类型可以是根据社交关系来划 分的，也可以是根据交互行为的累计时长来划分的，还可以是根据目标用户与第一相邻用户成为好友的时间(生成第一次交互行为的时间)来划分的，等等。
例如，当根据社交关系来划分时，根据目标用户与第一相邻用户之间的交互行为，将目标用户的朋友划分在第一个第一关系类型中，将目标用户的家人划分在第二个第一关系类型中，将目标用户的同事划分在第三个第一关系类型中。
S202：根据目标用户的交互向量和相邻用户的交互向量集合，得到目标用户的社交特征信息，社交特征信息是根据目标用户在K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的。
目标用户的交互向量用于表示目标用户在交互行为中的特征，在一个实施例中，目标用户的交互向量可以是基于自身携带的用户特征信息得到的，其中，用户特征信息包括目标用户所在的各种关系中其他用户的信息，也可以是基于目标用户的第一相邻用户的交互向量得到的，还可以是基于目标用户的第一相邻用户至第S相邻用户的交互向量得到的，S为正整数。可以理解的是，S与目标用户的交互向量中携带的用户特征信息的信息量成正比。
类似地，相邻用户的交互向量集合中，各个目标用户的第一相邻用户的交互向量，可以是基于第一相邻用户自身携带的用户特征信息得到的，也可以是基于该第一相邻用户的第一相邻用户的交互向量得到的，还可以是基于该第一相邻用户的第一相邻用户至第S相邻用户的交互向量得到的。
目标用户的社交特征信息具体可以是一个特征向量，也可以是一个特征矩阵。在一种实施方式中，根据目标用户的交互向量和相邻用户的交互向量集合，计算得到K个第一关系类型中，每个第一关系类型对应的关系特征信息；在得到K个第一关系类型中，每个第一关系类型对应的关系特征信息后，通过K个第一关系类型分别对应的K个关系特征信息，对目标用户进行特征分析，得到目标用户的社交特征信息。也就是说，目标用户的社交特征信息是由K个第一关系类型分别对应的K个关系特征信息共同表示的，即，目标用户的社交特征信息是基于目标用户与相邻用户的各个关系类型得到的。
S203：获取多媒体资源集合，并根据目标用户的社交特征信息，在目标用户的终端设备上展示多媒体资源集合中的第一多媒体资源。
其中，多媒体资源集合可以是预设的，也可以是多媒体资源平台根据数据库中的多媒体资源实时更新得到的。
在一种实施方式中，多媒体资源集合中的每个多媒体资源的资源特征信息，是通过访问过该多媒体资源的用户的交互向量得到的。例如，选取50个点击过该多媒体资源的用户的交互向量，并通过这50个交互向量计算该多媒体资源的资源特征信息。
根据目标用户的社交特征信息和多媒体资源集合中各个多媒体资源的资源特征信息，在目标用户的终端设备上展示多媒体资源集合中的第一多媒体资源，第一多媒体资源是多媒体资源集合中与目标用户的社交特征信息匹配度高于匹配度阈值的一个或多个多媒体资源，即目标用户最可能感兴趣的多媒体资源。
本申请实施例中，获取多个用户的终端设备之间的交互数据；根据交互数据，构建社交关系网络图；根据社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，根据目标用户的交互向量和相邻用户的交互向量集合，得到目标用户的社交特征信息，社交特征信息是根据目标用户与相邻用户的K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的，获取多媒体资源集合，并根据目标用户的社交特征信息，向目标用户的终端设备发送多媒体资源集合中的第一多媒体资源。
可见，由于用户的社交网络中含有大量具有相同兴趣的好友，如篮球、羽毛球、游泳等，并且用户通常与好友有着相似的圈子，如：金融圈、学生圈、科研圈、消费等级等，通过目标用户与目标用户的相邻用户的关系类型，对用户的社交关系进行分离,从而可以从不同的圈子与兴趣群的角度，挖掘目标用户的兴趣点，这样，能够更加深入与全面地对用户的交互行为进行表示。基于这种 社交关系的分类，进而对用户的社交特征进行聚合、对多媒体资源进行预测，使得在用户浏览行为不足的条件下，也可以更好地利用社交信息，更准确地提供多媒体资源。例如，用户加入某个短视频平台不久，在浏览短视频的初期，就能够获得非常精准的短视频推荐；对于业务处于起步阶段的短视频平台来说，提升了该平台处理推荐视频的精准率。
请参阅图3，图3为本申请实施例提供的另一种多媒体资源的处理方法的流程示意图。本申请实施例的所述方法应用于电子设备，该电子设备例如可以是上述提及的某些用户所使用的终端设备，如上述图1中的终端设备101，也可以是某些具有特殊功能的服务器，如上述图1中的服务器102。所述方法包括如下步骤。
S301：获取社交关系信息集合，并根据社交关系信息集合生成社交关系网络图。社交关系信息集合包括用户信息集合和关系信息集合。
在一种实施方式中，用户信息集合包括了各个用户的用户信息，例如，每个用户的性别、年龄、爱好等。电子设备根据用户信息集合，生成N个网络节点，N个网络节点中每个网络节点对应一个用户，且每个网络节点中携带有与该网络节点所对应用户的用户信息，N为正整数；各个网络节点之间的连边是根据各个网络节点对应的用户之间的交互行为确定的。
关系信息集合包括了两两用户所构成的社交关系相关的信息，例如，如交互行为记录、分组信息、标签信息等。若关系信息集合指示N个网络节点中的第一网络节点对应的用户和第二网络节点的用户存在交互行为(如聊天、转账、社交圈互动等)，则根据交互行为，生成第一网络节点和第二网络节点的连边信息，得到社交关系网络图。
进一步地，第一网络节点和第二网络节点的连边信息包括连边权重。电子设备可以根据关系信息集合，确定社交关系网络图中各个边的连边权重。具体地，根据第一网络节点和第二网络节点之间的关联度，确定连边权重，连边权重与关联度成正比，关联度是根据第一网络节点和第二网络节点在目标时间段内的交互信息确定的，所述交互信息包括：累积交互次数、累积交互时长、交互频率、交互内容中的至少一项。
例如，用户A和用户B在一周内的累积聊天(交互)时长为10小时，用户A和用户C在一周内的累积聊天时长为20分钟，则用户A和用户B之间的连边权重大于用户A和用户C之间的连边权重。
在一个实施例中，首先定义用户社交关系网络图G＝(A,E)。全体用户数量为N，A是用户关联矩阵，E是用户特征信息。在刻画用户社交关系时，需要多种信息，如：用户聊天数量，用户聊天频率，用户朋友圈互动评率等行为特征。电子设备基于用户的这些历史行为，将各个用户对应的网络节点，连接成用户社交关系网络图。用户间的连边权重由用户关联度决定。如果两个用户间有较多交互行为，则两个用户连边权重较高；如果两个用户间交互行为较少，则两个用户连边权重较低；如果两个用户间没有交互行为，则两个用户间没有连线。其中，衡量用户交互行为，可以通过用户交互行为次数、交互行为累计时长、交互论述、交互频率排序、最近一星期交互天数等变量共同决定。
图4a为本申请实施例提供的一种用户社交关系网络图的示意图。如图4a所示，设网络节点u1和u2对应的用户的交互次数较多，网络节点u1和u3对应的用户的交互次数较少，则网络节点u1和u2之间的连边权重比网络节点u1和u3之间的连边权重高。设网络节点u1对应的用户与除网络节点u2和u3外的其他网络节点对应的用户无交互，则u1与其他网络节点无连线。
此外，为了更好地描述用户i和用户j之间的关联度，可以将两个用户的交互次数记为c ij,则用户i和用户j之间的关联关系可以表示为：log(1+c ij)。也就是说，在用户关联矩阵A中，A ij＝log(1+c ij)，i，j均为正整数，i不等于j，且i，j均小于等于N。
S302：根据社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合。
在得到社交关系网络图后，本步骤S302中，基于社交关系网络图，对用户的社交关系进行分离，分别得到目标用户的交互向量，以及各个相邻用户的交互向量构成的集合。
具体地，电子设备根据社交关系网络图，得到N个网络节点对应的N个用户的交互向量，其目的在于通过向量来描述用户在社交关系网络中的交互行为，使得社交关系亲近的用户的向量表示较为相近；相应地，社交关系疏远的用户，向量表示差异较大。
具体地，基于社交关系网络图，以目标用户对应的目标网络节点为起点，在社交关系网络图中进行游走，得到M条轨迹，每条轨迹的步长为P；其中，M，P均为正整数；根据M条轨迹中携带的用户信息，得到目标用户的交互向量。
这样，若目标用户为用户i，用户i的第m条轨迹所携带的用户信息为U i,m，m＝1,…,M，那么，目标用户的交互向量x i可以表示为：
x i＝{U i,1,U i,2,…,U i,M}          (1)
其中，从第i个网络节点游走至第j个网络节点的概率，与第i个网络节点和第j个网络节点的连边信息中的连边权重成正比，i，j均为正整数，i不等于j，且i，j均小于等于N。也就是说，A ij越大，则从网络节点i游走至网络节点j的概率越大。
在另一个实施例中，电子设备通过向量化嵌入等方法对用户进行交互向量的表示，常用的向量化嵌入方法包括Node2Vec节点嵌入等无监督用户嵌入方法。
以Node2Vec节点嵌入方法为例，从社交关系网络图中的目标网络节点出发，游走多条轨迹；随后，将全部游走出的轨迹作为语料库，输入到word2vec词向量嵌入算法模型。通过word2vec词向量嵌入算法模型，对语料库进行处理，得到目标网络节点对应的目标用户的交互向量。
由于社交关系网络图中，不同用户对应的网络节点之间的连边权重不同，因此，在进行向量化嵌入的过程中，可以考虑到权重的影响，使用携带权重的游走方式。这样，从网络节点i游走至网络节点j的概率，与A ij成正比。
按照上述方法，电子设备可以得到社交关系网络图中所有节点对应的用户的交互向量构成的矩阵，记为X，
X＝{x 1,x 2,…,x N}           (2)
其中，x i表示用户i的交互向量。
在一种实施方式中，在得到社交关系网络图中所有节点对应的用户的交互向量构成的矩阵X后，电子设备可以从矩阵X中，抽取出目标用户的交互向量和相邻用户的交互向量集合。
S303：根据目标用户的交互向量和相邻用户的交互向量集合，得到K个第一关系类型中，每个第一关系类型对应的关系特征信息。
图4b为本申请实施例提供的一种基于社交关系网络图的图卷积模型的示意图。如图4b所示，V0为目标用户对应的目标网络节点，V1-V8为目标用户的第一相邻用户对应的网络节点。
在一种实施方式中，V1-V8有相同的聚合权重，以及相同的映射函数。也就是说，不考虑V1-V8与V0的关联度，认为V1-V8对V0的影响力相同。例如，某个群组或者关系群是目标用户很少浏览的对象，其中的各个相邻用户与目标用户都是弱连接的关系，那么，此时认为各个相邻用户与目标用户的连边权重相同。
在另一种实施方式中，由于V1-V8是V0通过不同方式建立社交关系的用户，例如，网络节点V1对应的用户是目标用户(V0)的家人，网络节点V5对应的用户是目标用户(V0)的朋友，网络节点V8是目标用户(V0)的客户。因此，各个相邻用户对目标用户的影响力不相同，其中，关系越亲密的人对目标用户影响力更高，也就是，关联度更高。对此，需要对不同用户进行分类，例如将用户按照建立连接的原因、关系类型、建立社交关系的累积时长、交互频率等因素，分成多类。
实践发现，在实际应用中，往往难以直接获取大量用户之间社交关系形成的原因。因此，本申请实施例通过一个类似于聚类的方法，来实现这一步骤S303。
在一个实施例中，电子设备按照预设规则，将目标用户的第一相邻用户划分为K个第一关系类型，例如，预先设置家人、朋友、客户等K个第一关系类型，获取K个第一关系类型中第h个第一关系类型的特征参数集合，特征参数集合包括第h个第一关系类型的权重矩阵和第h个第一关系类型的偏置向量。其中，h为正整数，h＝1,…,K。
具体地，首先按照预设规则，初始化第h个第一关系类型的权重矩阵W h和第h个第一关系类型的偏置向量b h。例如，对第h个第一关系类型的权重矩阵W h和第h个第一关系类型的偏置向量b h进行随机初始化。然后，在训练过程中，通过梯度下降法，对第h个第一关系类型的权重矩阵W h和第h个第一关系类型的偏置向量b h进行更新，最终得到第h个第一关系类型更新后的权重矩阵W h和第h个第一关系类型的偏置向量b h。
同理，电子设备可以基于上述方法得到各个第一关系类型的权重矩阵和偏置向量。
进一步地，通过第h个第一关系类型的特征参数集合，计算目标用户在第h个第一关系类型下的目标用户中间特征(即目标用户的隐表示)，并计算目标用户的各个第一相邻用户的在第h个第一关系类型下的相邻用户中间特征(即第一相邻用户的隐表示)。
这里，目标用户中间特征和相邻用户中间特征中的每个中间特征，可以根据用户的交互向量x i、权重矩阵W h以及偏置向量b h，进行计算。
具体地，用户i在第h个第一关系类型中的隐表示z i,h可以表示为：
Figure PCTCN2022095887-appb-000001
其中，||x|| 2表示计算x的模长，除以模长的操作是为了摆脱向量长度对分类的影响；σ(x)为激活函数(如sigmoid函数、tanh函数、Relu函数等)，
Figure PCTCN2022095887-appb-000002
为第h个第一关系类型的权重矩阵W h的转置矩阵，x i为用户i的交互向量(从步骤S302中的矩阵X中可以得到)，b h为第h个第一关系类型的偏置向量。
这样，基于上述公式(3)，电子设备可以计算得到目标用户和目标用户的各个第一相邻用户的隐表示。
在一个实施例中，根据目标用户中间特征和各个相邻用户中间特征，得到第h个第一关系类型的关系特征信息。
也就是，在计算得到目标用户的隐表示和目标用户的第一相邻用户的隐表示后，电子设备根据目标用户的隐表示和目标用户的第一相邻用户的隐表示，得到第h个第一关系类型的关系特征信息c h，具体可以表示为：
Figure PCTCN2022095887-appb-000003
其中，网络节点u对应图4b中的V0(即目标用户对应的网络节点)，(v|(u,v)∈G)对应图4b中的V1-V8(即目标用户的第一相邻用户对应的网络节点)，p v,h用于表示用户v被分配到第h个第一关系类型的概率，p v,h≥0，且
Figure PCTCN2022095887-appb-000004
z u,h为目标用户在第h个第一关系类型中的隐表示，z v,h为目标用户的第一相邻用户在第h个第一关系类型中的隐表示。
其中，p v,h的值是根据第一关系类型的数量决定的。在一种具体实施方式中
Figure PCTCN2022095887-appb-000005
例如，设用户A的第一关系类型的数量为5(即用户A的社交关系被划分为5类)，则
Figure PCTCN2022095887-appb-000006
这样，基于上述公式(4)，电子设备可以计算得到K个第一关系类型的关系特征信息。
在另一个实施例中，采用迭代的方式，根据目标用户中间特征和各个相邻用户中间特征，得到第h个第一关系类型的关系特征信息。
具体地，第h个第一关系类型的关系特征信息是第h个第一关系类型在第T次迭代时得到的关系特征信息，T为正整数。电子设备采用迭代的方式，获取目标用户的各个第一相邻用户在第t次迭代时，被划分为第h个第一关系类型的目标概率
Figure PCTCN2022095887-appb-000007
t为正整数，且t小于T；然后电子设备根据目标概率
Figure PCTCN2022095887-appb-000008
和目标用户的各个相邻用户中间特征(即第一相邻用户的隐表示(z v,h))，计算目标用户的第一相邻用户的聚合特征
Figure PCTCN2022095887-appb-000009
对目标用户中间特征(即目标用户的隐表示(z u,h))和目标用户的第一相邻用户的聚合特征进行运算处理，得到第h个第一关系类型在第t+1次迭代时得到的关系特征信息。
具体地，目标概率
Figure PCTCN2022095887-appb-000010
可以表示为：
Figure PCTCN2022095887-appb-000011
其中，指数函数exp(x)表示计算x的指数，
Figure PCTCN2022095887-appb-000012
为z v,h的转置矩阵，
Figure PCTCN2022095887-appb-000013
表示在第t次迭代时得到的第h个第一关系类型的关系特征信息。
进一步地，
Figure PCTCN2022095887-appb-000014
可以表示为：
Figure PCTCN2022095887-appb-000015
其中，网络节点u对应图4b中的V0(即目标用户对应的网络节点)，(v|(u,v)∈G)对应图4b中的V1-V8(即目标用户的第一相邻用户对应的网络节点)，
Figure PCTCN2022095887-appb-000016
用于表示在第t-1次迭代时，用户v被分配到第h个第一关系类型的概率，z u,h为目标用户在第h个第一关系类型中的隐表示，z v,h为目标用户的第一相邻用户在第h个第一关系类型中的隐表示。其中，用户v被分配到第h个第一关系类型的概率的初始值
Figure PCTCN2022095887-appb-000017
是根据第一关系类型的数量决定的。
基于上述公式(5)和公式(6)进行迭代运算，电子设备可以计算出每个第一关系类型在第T次迭代时得到的关系特征信息，其中，第T次迭代为最后一次运算，这样就得到了K个第一关系类型的关系特征信息。
实践发现，通过T次迭代，对用户v被分配到第h个第一关系类型的概率得到优化。这种基于交替更新的归类方法，可以区分不同第一关系类型对于用户v的影响，实现了对关系的智能聚合，能有效地提高多媒体资源推荐的准确率。
S304：通过K个第一关系类型分别对应的K个关系特征信息，对目标用户进行特征分析，得到目标用户的社交特征信息。
在一种实施方式中，社交特征信息是K个关系特征信息的集合。
具体地，电子设备将每个第一关系类型在第T次迭代时得到的关系特征信息，确定为该第一关系类型的关系特征信息，即
Figure PCTCN2022095887-appb-000018
进而得到目标用户的社交特征信息y u＝[c 1,c 2,…,c K]。
在另一种实施方式中，电子设备获取目标用户的第一相邻用户的第一特征信息集合，和目标用户的第二相邻用户的第二特征信息集合，其中，第二相邻用户指与第一相邻用户存在交互行为、且与目标用户不存在交互行为的用户，第一特征信息集合包括目标用户的第一相邻用户的多个第二关系类型中，每个第二关系类型对应的关系特征信息(如[c′ 1,c′ 2,…,c′ R])，第二特征信息集合包括目标用户的第二相邻用户的多个第三关系类型中，每个第三关系类型对应的关系特征信息(如[c″ 1,c″ 2,…,c″ S])；R，S为正整数，且R，S，K可以相同也可以不同。
电子设备获取与目标用户的第一相邻用户的第一特征信息集合，和目标用户的第二相邻用户的第二特征信息集合的具体实施方式，可参考步骤S301-步骤S303，在此不再赘述。
通过第一特征信息集合，可以得到目标用户的第一相邻用户的社交特征信息
Figure PCTCN2022095887-appb-000019
同理，通过第二特征信息集合，可以得到目标用户的第二相邻用户的社交特征信息
Figure PCTCN2022095887-appb-000020
Figure PCTCN2022095887-appb-000021
以此类推，通过第L特征信息集合，可以得到目标用户的第L相邻用户的社交特征信息
Figure PCTCN2022095887-appb-000022
将目标用户的K个第一关系类型对应的关系特征信息、第一特征信息集合至第L特征信息集合，输入到关系预测模型中，得到关系预测模型输出的预测结果。例如，L＝2。
然后，对预测结果进行过拟合处理，得到目标用户的社交特征信息。
其中，关系预测模型包括L层图卷积网络层，L为正整数。上述图4b所示的图卷积模型表征了一层图卷积网络层，关系预测模型一共包括L个级联的图卷积网络层，其中，第g层图卷积网络层的输入数据包括：对第g-1层图卷积网络层的输出数据进行过拟合处理后得到的数据。
具体地，关系预测模型第l层的输入为目标用户在第l-1层的处理结果
Figure PCTCN2022095887-appb-000023
以及目标用户的 第一相邻用户在第l-1层的处理结果集合
Figure PCTCN2022095887-appb-000024
l为正整数，且l小于等于L，采用关系预测模型的第l层对输入数据进行处理可以表示为：
Figure PCTCN2022095887-appb-000025
进一步地，对关系预测模型的第l层的输出数据进行过拟合处理，得到目标用户在第l层的处理结果
Figure PCTCN2022095887-appb-000026
Figure PCTCN2022095887-appb-000027
其中，f (l)(x)表示通过关系预测模型的第l层对x进行处理，dropout(x)表示对x进行过拟合处理，
Figure PCTCN2022095887-appb-000028
的值被初始化为x u。目标用户在第L层的处理结果
Figure PCTCN2022095887-appb-000029
可以表示为：
Figure PCTCN2022095887-appb-000030
需要说明的是，U u与y u的区别在于，y u是通过目标用户的交互向量和目标用户的第一相邻用户的交互向量得到的，U u是通过目标用户的交互向量和目标用户的第一相邻用户的交互向量至第L相邻用户的交互向量得到的。因此，U u相对于y u涵盖了更多的特征信息。
S305：获取多媒体资源集合，并根据目标用户的社交特征信息，在目标用户的终端设备上展示多媒体资源集合中的第一多媒体资源。
在一种实施方式中，多媒体资源集合中包括第二多媒体资源的资源特征信息。电子设备获取第二多媒体资源的访问记录，访问记录中包括访问过第二多媒体资源的Q个用户的用户标识，Q为正整数；根据Q个用户的用户标识，获取Q个用户的交互向量(如从步骤S302中的矩阵X中获取)；对Q个用户的交互向量进行融合处理，如对Q个用户的交互向量进行叠加，并对叠加后的结果进行均值池化(mean pooling)处理，得到第二多媒体资源的资源特征信息，其中，多媒体资源集合中包括第二多媒体资源的资源特征信息。
具体地，假设访问第二多媒体资源的用户列表为(u 1,u 2,u 3,u 4,…u H)，其中H是访问过该视频的所有用户的总量。若第二多媒体资源的资源特征信息表示为i m，则：
Figure PCTCN2022095887-appb-000031
其中，
Figure PCTCN2022095887-appb-000032
为用户u h的交互向量。
类似地，电子设备可以基于上述公式(10)，得到多媒体资源集合中所有多媒体资源的资源特征信息，这些资源特征信息可以记录在矩阵I中，I＝{i 1,i 2,…,i N}。
进一步地，电子设备根据目标用户的社交特征信息，和多媒体资源集合中各个多媒体资源的资源特征信息，计算目标用户与多媒体资源集合中各个多媒体资源的匹配度，其中，第一多媒体资源是多媒体资源集合中与目标用户匹配度最高的多媒体资源。
在一种实施方式中，电子设备可以通过多层感知机(Muti-Layer Percetron，MLP)，预测每个第一关系类型与多媒体资源i的关系，然后通过注意力机制，综合考虑不同第一关系类型的偏好，如目标用户的同事喜欢看古装剧，目标用户的家人喜欢看都市剧，最终得到目标用户u和多媒体资源i的预测结果。
具体地，电子设备将多媒体资源集合中第二多媒体资源的资源特征信息与目标用户的社交特征信息进行拼接，得到拼接特征集合，即，将第二多媒体资源的资源特征信息，与[c u,1,c u,2,…,c u,K]中的每个第一关系类型对应的关系特征信息进行拼接，得到K个拼接特征。
然后，采用多层感知机，对拼接特征集合中的各个拼接特征进行处理，得到K个第一关系类型中，每个第一关系类型与第二多媒体资源的关系向量。
具体地，目标用户的第k个第一关系类型与多媒体资源i的关系向量可以表示为：
r u,i,k＝MLP 1(c u,k||I i)          (11)
其中，x||y表示对向量x和向量y进行拼接，MLP 1(x)表示采用第一多层感知机对x进行处理。
基于上述公式(11)，智能设备可以通过第一多层感知机，对拼接特征集合中的各个拼接特征 进行处理，得到各个第一关系类型与各个多媒体资源的关系向量。
在得到各个第一关系类型与第二多媒体资源的关系向量后，电子设备根据每个第一关系类型与第二多媒体资源的关系向量，计算该第一关系类型对应的权重.
具体地，目标用户的第k个第一关系类型与多媒体资源i的权重可以表示为：
Figure PCTCN2022095887-appb-000033
其中，指数函数exp(x)表示计算x的指数，σ(x)为激活函数(如sigmoid函数)，a T为注意力向量，a Tr u,i,k表示将关系向量与注意力向量进行点乘运算。基于上述公式(12)，电子设备可以计算得到目标用户的各个第一关系类型与各个多媒体资源间的权重。
在得到目标用户的每个第一关系类型与第二多媒体资源间的权重后，电子设备根据K个第一关系类型中，每个第一关系类型与第二多媒体资源的关系向量，以及每个第一关系类型对应的权重，得到目标用户与第二多媒体资源的匹配度.
具体地，目标用户与多媒体资源i的匹配度可以表示为：
Figure PCTCN2022095887-appb-000034
其中，MLP 2(x)表示采用第二多层感知机对x进行处理。基于上述公式(13)，电子设备可以通过第二多层感知机，得到目标用户与多媒体资源集合中各个多媒体资源的匹配度。
进一步地，电子设备按照匹配度由高至低的顺序，对多媒体资源集合中的多媒体资源进行排序，并将排列在目标位置前的(一个或多个)多媒体资源确定为第一多媒体资源。然后向目标用户推荐第一多媒体资源。当电子设备为服务器时，服务器将第一多媒体资源发送给目标用户的终端设备进行展示；当电子设备为终端设备时，将确定出的第一多媒体资源展示在界面上，如图1中的页面103所示。
在另一种实施方式中，在进行多媒体资源推荐之前，电子设备可以采用训练数据，对公式(3)-公式(13)中的参数进行优化，即，将标注数据与通过公式(3)-公式(13)计算得到的预测数据进行比对，并通过损失函数，调整公式(3)-公式(13)中的参数，以降低标注数据和预测数据间的差异，直至损失函数收敛。
在训练完成后，电子设备获取到全部用户的交互向量。当检测到目标用户u的多媒体资源获取请求时，通过执行上述步骤S301-步骤S305，比较目标用户u与多媒体资源集合中各个多媒体资源的相似度，随后向目标用户u推荐满足推荐条件的多媒体资源。
本申请实施例在图2实施例的基础上，通过目标用户和目标用户的第一相邻用户的隐表示，得到第一关系类型中，每个第一关系类型对应的关系特征信息，进而得到目标用户的社交特征信息；通过观看过多媒体资源的用户的交互向量，得到多媒体资源的资源特征信息，进而通过多层感知机MLP，确定目标用户的社交特征信息和多媒体资源的资源特征信息之间的匹配度，从而向目标用户推荐多媒体资源。对此，通过多层感知机MLP，可以预测每一个团体向量与视频i的关系，然后通过注意力机制，综合考虑不同团体向量的兴趣，最终得到用户u和视频i的预测结果。这种基于注意力机制的用户与视频匹配度的预测方式，可以更好地从不同社交关系的角度来建模用户u对视频i的兴趣，使得对多媒体资源的推荐，可以脱离于历史浏览数据，同时保证推荐的精准度。
上述详细阐述了本申请实施例的方法，为了便于更好地实施本申请实施例的上述方案，相应地，下面提供了本申请实施例的装置。
请参见图5，图5为本申请实施例提供的一种多媒体资源的处理装置的结构示意图，多媒体资源的处理装置500包括获取单元501和处理单元502，该装置可以搭载在电子设备上，电子设备可以包括终端设备、服务器。图5所示的多媒体资源的处理装置可以用于执行上述图2和图3所描述的方法实施例中的部分或全部功能。其中，各个单元的详细描述如下：
获取单元501，用于获取多个用户的终端设备之间的交互数据；根据所述交互数据，构建社交关系网络图；根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
处理单元502，用于根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
在一个实施例中，所述处理单元502，具体用于：
根据所述目标用户的交互向量和所述相邻用户的交互向量集合，计算得到所述K个第一关系类型中，每个第一关系类型对应的关系特征信息；
通过所述K个第一关系类型分别对应的K个关系特征信息，对所述目标用户进行特征分析，得到所述社交特征信息。
在一个实施例中，所述处理单元502，具体用于：
获取所述K个第一关系类型中第h个第一关系类型的特征参数集合，所述特征参数集合包括所述第h个第一关系类型的权重矩阵和所述第h个第一关系类型的偏置向量，其中，h为正整数，且h小于K；
通过所述第h个第一关系类型的特征参数集合，计算所述目标用户在所述第h个第一关系类型下的目标用户中间特征，并计算所述目标用户的各个第一相邻用户在所述第h个第一关系类型下的相邻用户中间特征；
根据所述目标用户中间特征和各个相邻用户中间特征，得到所述第h个第一关系类型的关系特征信息。
在一个实施例中，所述第h个第一关系类型的关系特征信息是所述第h个第一关系类型在第T次迭代时得到的关系特征信息，T为正整数；所述处理单元502，具体用于：
采用迭代的方式，获取所述目标用户的各个第一相邻用户在第t次迭代时，被划分为所述第h个第一关系类型的目标概率，t为正整数，且t小于T；
根据所述目标概率和各个相邻用户中间特征，计算第一相邻用户的聚合特征；
对所述目标用户中间特征和所述聚合特征进行运算处理，得到所述第h个第一关系类型在第t+1次迭代时得到的关系特征信息。
在一个实施例中，所述处理单元502，具体用于：
获取所述第一相邻用户的第一特征信息集合，和所述目标用户的第二相邻用户的第二特征信息集合，其中，所述第二相邻用户指与所述第一相邻用户存在交互行为、且与所述目标用户不存在交互行为的用户，所述第一特征信息集合包括：所述第一相邻用户的多个第二关系类型中，每个第二关系类型对应的关系特征信息，所述第二特征信息集合包括：所述第二相邻用户的多个第三关系类型中，每个第三关系类型对应的关系特征信息；
将所述目标用户的K个第一关系类型对应的关系特征信息、所述第一特征信息集合以及所述第二特征信息集合，输入到关系预测模型中，得到所述关系预测模型输出的预测结果；
对所述预测结果进行过拟合处理，得到所述社交特征信息。
在一个实施例中，所述关系预测模型包括L层图卷积网络层，L为正整数，第g层图卷积网络层的输入数据包括：对第g-1层图卷积网络层的输出数据进行过拟合处理后得到的数据。
在一个实施例中，所述处理单元502，具体用于：
获取第二多媒体资源的访问记录，所述访问记录中包括访问过所述第二多媒体资源的Q个用户的用户标识，Q为正整数；
根据所述Q个用户的用户标识，获取所述Q个用户的交互向量；
对所述Q个用户的交互向量进行叠加，并对叠加后的结果进行均值池化处理，得到所述第二 多媒体资源的资源特征信息，其中，所述多媒体资源集合中包括所述第二多媒体资源的资源特征信息。
在一个实施例中，所述处理单元502，具体用于：
根据所述目标用户的社交特征信息，和所述多媒体资源集合中各个多媒体资源的资源特征信息，计算所述目标用户与所述多媒体资源集合中各个多媒体资源的匹配度，其中，所述第一多媒体资源是所述多媒体资源集合中与所述目标用户匹配度最高的多媒体资源。
在一个实施例中，所述处理单元502，具体用于：
将所述第二多媒体资源的所述资源特征信息与所述目标用户的所述社交特征信息进行拼接，得到拼接特征集合；
采用多层感知机，对所述拼接特征集合中的各个拼接特征进行处理，得到所述K个第一关系类型中，每个第一关系类型与所述第二多媒体资源的关系向量；
根据每个第一关系类型与所述第二多媒体资源的关系向量，计算该第一关系类型对应的权重；
根据所述K个第一关系类型中，每个第一关系类型与所述第二多媒体资源的关系向量，以及每个第一关系类型对应的权重，得到所述目标用户与所述第二多媒体资源的匹配度。
在一个实施例中，所述获取单元501，具体用于：
获取社交关系信息集合，所述社交关系信息集合包括用户信息集合和关系信息集合；
根据所述用户信息集合，生成N个网络节点，所述N个网络节点中每个网络节点对应一个用户，且每个网络节点中携带有与该网络节点所对应用户的用户信息，N为正整数；
若所述关系信息集合指示所述N个网络节点中的第一网络节点对应的用户和第二网络节点的用户存在交互行为，则根据所述交互行为，生成所述第一网络节点和所述第二网络节点的连边信息，得到所述社交关系网络图。
在一个实施例中，所述连边信息包括所述第一网络节点和所述第二网络节点的连边权重，所述连边权重是根据所述第一网络节点和所述第二网络节点之间的关联度确定的，所述连边权重与所述关联度成正比。
在一个实施例中，所述关联度是根据所述第一网络节点和所述第二网络节点在目标时间段内的交互信息确定的，所述交互信息包括：累积交互次数、累积交互时长、交互频率、交互内容中的至少一项。
在一个实施例中，所述获取单元501，具体用于：
以所述目标用户对应的目标网络节点为起点，在所述社交关系网络图中进行游走，得到M条轨迹，每条轨迹的步长为P；其中，M，P均为正整数；
根据所述M条轨迹中携带的用户信息，得到所述目标用户的交互向量。
在一个实施例中，从第i个网络节点游走至第j个网络节点的概率，与所述第i个网络节点和所述第j个网络节点的连边权重成正比，i，j均为正整数，i不等于j，且i，j均小于等于N。
根据本申请的一个实施例，图2和图3所示的多媒体资源的处理方法所涉及的部分步骤可由图5所示的多媒体资源的处理装置中的各个单元来执行。例如，图2中所示的步骤S200和步骤S201可由图5所示的获取单元501执行，步骤S202和步骤S203可由图5所示的处理单元502执行。图3中所示的步骤S301和步骤S302可由图5所示的获取单元501执行，步骤S303-步骤S305可由图5所示的处理单元502执行。
图5所示的多媒体资源的处理装置中的各个单元可以分别或全部合并为一个或若干个另外的单元来构成，或者其中的某个(些)单元还可以再拆分为功能上更小的多个单元来构成，这可以实现同样的操作，而不影响本申请的实施例的技术效果的实现。上述单元是基于逻辑功能划分的，在实际应用中，一个单元的功能也可以由多个单元来实现，或者多个单元的功能由一个单元实现。在本申请的其它实施例中，多媒体资源的处理装置也可以包括其它单元，在实际应用中，这些功能也可以由其它单元协助实现，并且可以由多个单元协作实现。
根据本申请的另一个实施例，可以通过在包括中央处理单元(CPU)、随机存取存储介质(RAM)、 只读存储介质(ROM)等处理元件和存储元件的例如计算机的通用计算装置上运行能够执行如图2和图3中所示的相应方法所涉及的各步骤的计算机程序(包括程序代码)，来构造如图5中所示的多媒体资源的处理装置，以及来实现本申请实施例的多媒体资源的处理方法。所述计算机程序可以记载于例如计算机可读记录介质上，并通过计算机可读记录介质装载于上述计算装置中，并在其中运行。
基于同一发明构思，本申请实施例中提供的多媒体资源的处理装置解决问题的原理与有益效果与本申请方法实施例中多媒体资源的处理装置解决问题的原理和有益效果相似，可以参见方法的实施的原理和有益效果，为简洁描述，在这里不再赘述。
请参阅图6，图6为本申请实施例提供的一种电子设备的结构示意图，所述电子设备600至少包括处理器601、通信接口602和存储器603。其中，处理器601、通信接口602和存储器603可通过总线或其他方式连接。其中，处理器601(或称中央处理器(Central Processing Unit，CPU))是终端的计算核心以及控制核心，其可以解析终端内的各类指令以及处理终端的各类数据，例如：CPU可以用于解析用户向终端所发送的开关机指令，并控制终端进行开关机操作；再如：CPU可以在终端内部结构之间传输各类交互数据，等等。通信接口602可选的可以包括标准的有线接口、无线接口(如WI-FI、移动通信接口等)，受处理器601的控制可以用于收发数据；通信接口602还可以用于终端内部数据的传输以及交互。存储器603(Memory)是终端中的记忆设备，用于存放程序和数据。可以理解的是，此处的存储器603既可以包括终端的内置存储器，当然也可以包括终端所支持的扩展存储器。存储器603提供存储空间，该存储空间存储了终端的操作系统，可包括但不限于：Android系统、iOS系统、Windows Phone系统等等，本申请对此并不作限定。
在本申请实施例中，处理器601通过运行存储器603中的可执行程序代码，用于执行如下操作：
通过通信接口602获取多个用户的终端设备之间的交互数据；
根据所述交互数据，构建社交关系网络图；
根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；
获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
基于同一发明构思，本申请实施例中提供的电子设备解决问题的原理与有益效果与本申请方法实施例中多媒体资源的处理方法解决问题的原理和有益效果相似，可以参见方法的实施的原理和有益效果，为简洁描述，在这里不再赘述。
本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质中存储有一条或多条指令，所述一条或多条指令适于由处理器加载并执行上述方法实施例所述的多媒体资源的处理方法。
本申请实施例还提供一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上述方法实施例所述的多媒体资源的处理方法。
本申请实施例还提供一种计算机程序产品或计算机程序，该计算机程序产品或计算机程序包括计算机指令，该计算机指令存储在计算机可读存储介质中。计算机设备的处理器从计算机可读存储介质读取该计算机指令，处理器执行该计算机指令，使得该计算机设备执行上述多媒体资源的处理方法。
本申请实施例方法中的步骤可以根据实际需要进行顺序调整、合并和删减。
本申请实施例装置中的模块可以根据实际需要进行合并、划分和删减。
本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序可以存储于一计算机可读存储介质中，可读存储介质可以包括：闪存盘、只读存储器(Read-Only Memory，ROM)、随机存取器(Random Access Memory，RAM)、磁盘或光盘等。
以上所揭露的仅为本申请一种较佳实施例而已，当然不能以此来限定本申请之权利范围，本领域普通技术人员可以理解实现上述实施例的全部或部分流程，并依本申请权利要求所作的等同变化，仍属于发明所涵盖的范围。

Claims (20)

  1. 一种多媒体资源的处理方法，由电子设备执行，包括：
    获取多个用户的终端设备之间的交互数据；
    根据所述交互数据，构建社交关系网络图；
    根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
    根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；及，
    获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
  2. 如权利要求1所述的方法，其中，所述根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，包括：
    根据所述目标用户的交互向量和所述相邻用户的交互向量集合，计算得到所述K个第一关系类型中，每个第一关系类型对应的关系特征信息；
    通过所述K个第一关系类型分别对应的K个关系特征信息，对所述目标用户进行特征分析，得到所述社交特征信息。
  3. 如权利要求2所述的方法，其中，所述根据所述目标用户的交互向量和所述相邻用户的交互向量集合，计算得到所述K个第一关系类型中，每个第一关系类型对应的关系特征信息，包括：
    获取所述K个第一关系类型中第h个第一关系类型的特征参数集合，所述特征参数集合包括所述第h个第一关系类型的权重矩阵和所述第h个第一关系类型的偏置向量，其中，h为正整数，且h小于K；
    通过所述第h个第一关系类型的特征参数集合，计算所述目标用户在所述第h个第一关系类型下的目标用户中间特征，并计算所述目标用户的各个第一相邻用户在所述第h个第一关系类型下的相邻用户中间特征；
    根据所述目标用户中间特征和各个相邻用户中间特征，得到所述第h个第一关系类型的关系特征信息。
  4. 如权利要求3所述的方法，其中，所述第h个第一关系类型的关系特征信息是所述第h个第一关系类型在第T次迭代时得到的关系特征信息，T为正整数；所述根据所述目标用户中间特征和各个相邻用户中间特征，得到所述第h个第一关系类型的关系特征信息，包括：
    采用迭代的方式，获取所述目标用户的各个第一相邻用户在第t次迭代时，被划分为所述第h个第一关系类型的目标概率，t为正整数，且t小于T；
    根据所述目标概率和各个相邻用户中间特征，计算第一相邻用户的聚合特征；
    对所述目标用户中间特征和所述聚合特征进行运算处理，得到所述第h个第一关系类型在第t+1次迭代时得到的关系特征信息。
  5. 如权利要求2所述的方法，其中，所述通过所述K个第一关系类型分别对应的K个关系特征信息，对所述目标用户进行特征分析，得到所述社交特征信息，包括：
    获取所述第一相邻用户的第一特征信息集合，和所述目标用户的第二相邻用户的第二特征信息集合，其中，所述第二相邻用户指与所述第一相邻用户存在交互行为、且与所述目标用户不存在 交互行为的用户，所述第一特征信息集合包括：所述第一相邻用户的多个第二关系类型中，每个第二关系类型对应的关系特征信息，所述第二特征信息集合包括：所述第二相邻用户的多个第三关系类型中，每个第三关系类型对应的关系特征信息；
    将所述目标用户的K个第一关系类型对应的关系特征信息、所述第一特征信息集合以及所述第二特征信息集合，输入到关系预测模型中，得到所述关系预测模型输出的预测结果；
    对所述预测结果进行过拟合处理，得到所述社交特征信息。
  6. 如权利要求5所述的方法，其中，所述关系预测模型包括L层图卷积网络层，L为正整数，第g层图卷积网络层的输入数据包括：对第g-1层图卷积网络层的输出数据进行过拟合处理后得到的数据。
  7. 如权利要求1所述的方法，其中，所述获取多媒体资源集合，包括：
    获取第二多媒体资源的访问记录，所述访问记录中包括访问过所述第二多媒体资源的Q个用户的用户标识，Q为正整数；
    根据所述Q个用户的用户标识，获取所述Q个用户的交互向量；
    对所述Q个用户的交互向量进行叠加，并对叠加后的结果进行均值池化处理，得到所述第二多媒体资源的资源特征信息，其中，所述多媒体资源集合中包括所述第二多媒体资源的资源特征信息。
  8. 如权利要求7所述的方法，其中，所述根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源，包括：
    根据所述目标用户的社交特征信息，和所述多媒体资源集合中各个多媒体资源的资源特征信息，计算所述目标用户与所述多媒体资源集合中各个多媒体资源的匹配度，其中，所述第一多媒体资源是所述多媒体资源集合中与所述目标用户匹配度最高的多媒体资源。
  9. 如权利要求8所述的方法，其中，所述根据所述目标用户的社交特征信息，和所述多媒体资源集合中各个多媒体资源的资源特征信息，计算所述目标用户与所述多媒体资源集合中各个多媒体资源的匹配度，包括：
    将所述第二多媒体资源的所述资源特征信息与所述目标用户的所述社交特征信息进行拼接，得到拼接特征集合；
    采用多层感知机，对所述拼接特征集合中的各个拼接特征进行处理，得到所述K个第一关系类型中，每个第一关系类型与所述第二多媒体资源的关系向量；
    根据每个第一关系类型与所述第二多媒体资源的关系向量，计算该第一关系类型对应的权重；
    根据所述K个第一关系类型中，每个第一关系类型与所述第二多媒体资源的关系向量，以及每个第一关系类型对应的权重，得到所述目标用户与所述第二多媒体资源的匹配度。
  10. 如权利要求1所述的方法，其中，所述获取多个用户的终端设备之间的交互数据，包括：
    获取社交关系信息集合，所述社交关系信息集合包括用户信息集合和关系信息集合；
    所述根据所述交互数据，构建社交关系网络图，包括：
    根据所述用户信息集合，生成N个网络节点，所述N个网络节点中每个网络节点对应一个用户，且每个网络节点中携带有与该网络节点所对应用户的用户信息，N为正整数；
    若所述关系信息集合指示所述N个网络节点中的第一网络节点对应的用户和第二网络节点的用户存在交互行为，则根据所述交互行为，生成所述第一网络节点和所述第二网络节点的连边信息，得到所述社交关系网络图。
  11. 如权利要求10所述的方法，其中，所述连边信息包括所述第一网络节点和所述第二网络节点的连边权重，所述连边权重是根据所述第一网络节点和所述第二网络节点之间的关联度确定的，所述连边权重与所述关联度成正比。
  12. 如权利要求11所述的方法，其中，所述关联度是根据所述第一网络节点和所述第二网络节点在目标时间段内的交互信息确定的，所述交互信息包括：累积交互次数、累积交互时长、交互频率、交互内容中的至少一项。
  13. 如权利要求11所述的方法，其中，所述根据所述社交关系网络图，获取目标用户的交互向量，包括：
    以所述目标用户对应的目标网络节点为起点，在所述社交关系网络图中进行游走，得到M条轨迹，每条轨迹的步长为P；其中，M，P均为正整数；
    根据所述M条轨迹中携带的用户信息，得到所述目标用户的交互向量。
  14. 如权利要求13所述的方法，其中，从第i个网络节点游走至第j个网络节点的概率，与所述第i个网络节点和所述第j个网络节点的连边权重成正比，i，j均为正整数，i不等于j，且i，j均小于等于N。
  15. 一种多媒体资源的处理装置，包括：
    获取单元，用于获取多个用户的终端设备之间的交互数据；根据所述交互数据，构建社交关系网络图；根据所述社交关系网络图，获取目标用户的交互向量和相邻用户的交互向量集合，所述相邻用户的交互向量集合中包括所述目标用户的第一相邻用户的交互向量，所述目标用户与所述第一相邻用户之间的关系被划分为K个第一关系类型，K为正整数；
    处理单元，用于根据所述目标用户的交互向量和所述相邻用户的交互向量集合，得到所述目标用户的社交特征信息，所述社交特征信息是根据所述目标用户在所述K个第一关系类型中，各个第一关系类型对应的关系特征信息确定的；获取多媒体资源集合，并根据所述目标用户的所述社交特征信息，在所述目标用户的终端设备上展示所述多媒体资源集合中的第一多媒体资源。
  16. 如权利要求15所述的装置，其中，所述处理单元用于，根据所述目标用户的交互向量和所述相邻用户的交互向量集合，计算得到所述K个第一关系类型中，每个第一关系类型对应的关系特征信息；通过所述K个第一关系类型分别对应的K个关系特征信息，对所述目标用户进行特征分析，得到所述社交特征信息。
  17. 如权利要求16所述的装置，其中，所述处理单元用于，获取所述第一相邻用户的第一特征信息集合，和所述目标用户的第二相邻用户的第二特征信息集合，其中，所述第二相邻用户指与所述第一相邻用户存在交互行为、且与所述目标用户不存在交互行为的用户，所述第一特征信息集合包括：所述第一相邻用户的多个第二关系类型中，每个第二关系类型对应的关系特征信息，所述第二特征信息集合包括：所述第二相邻用户的多个第三关系类型中，每个第三关系类型对应的关系特征信息；将所述目标用户的K个第一关系类型对应的关系特征信息、所述第一特征信息集合以及所述第二特征信息集合，输入到关系预测模型中，得到所述关系预测模型输出的预测结果；对所述预测结果进行过拟合处理，得到所述社交特征信息。
  18. 如权利要求15所述的装置，其中，所述获取单元用于，获取第二多媒体资源的访问记录，所述访问记录中包括访问过所述第二多媒体资源的Q个用户的用户标识，Q为正整数；根据所述Q个用户的用户标识，获取所述Q个用户的交互向量；对所述Q个用户的交互向量进行叠加，并 对叠加后的结果进行均值池化处理，得到所述第二多媒体资源的资源特征信息，其中，所述多媒体资源集合中包括所述第二多媒体资源的资源特征信息。
  19. 一种电子设备，包括：存储装置和处理器；
    所述存储装置中存储有计算机程序；
    处理器，执行计算机程序，实现如权利要求1-14任一项所述的多媒体资源的处理方法。
  20. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时，如权利要求1-14任一项所述的多媒体资源的处理方法被实现。
PCT/CN2022/095887 2021-06-17 2022-05-30 多媒体资源的处理方法、装置、设备及存储介质 WO2022262561A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110675393.6 2021-06-17
CN202110675393 2021-06-17

Publications (1)

Publication Number Publication Date
WO2022262561A1 true WO2022262561A1 (zh) 2022-12-22

Family

ID=84464909

Family Applications (2)

Application Number Title Priority Date Filing Date
PCT/CN2022/095887 WO2022262561A1 (zh) 2021-06-17 2022-05-30 多媒体资源的处理方法、装置、设备及存储介质
PCT/CN2022/097098 WO2022262603A1 (zh) 2021-06-17 2022-06-06 多媒体资源的推荐方法、装置、设备、存储介质及计算机程序产品

Family Applications After (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/097098 WO2022262603A1 (zh) 2021-06-17 2022-06-06 多媒体资源的推荐方法、装置、设备、存储介质及计算机程序产品

Country Status (4)

Country Link
US (1) US20230273951A1 (zh)
JP (1) JP2024509966A (zh)
CN (1) CN115495598A (zh)
WO (2) WO2022262561A1 (zh)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130097056A1 (en) * 2011-10-13 2013-04-18 Xerox Corporation Methods and systems for recommending services based on an electronic social media trust model
CN105468723A (zh) * 2015-11-20 2016-04-06 小米科技有限责任公司 信息推荐方法及装置
CN105653535A (zh) * 2014-11-13 2016-06-08 中国科学院沈阳计算技术研究所有限公司 一种媒体资源推荐方法
CN106022938A (zh) * 2016-06-02 2016-10-12 北京奇艺世纪科技有限公司 社交网络的用户社团划分方法和装置
CN108280458A (zh) * 2017-01-05 2018-07-13 腾讯科技（深圳）有限公司 群体关系类型识别方法及装置
CN109544396A (zh) * 2019-01-10 2019-03-29 腾讯科技（深圳）有限公司 账号推荐方法、装置、服务器、终端及存储介质
CN110569437A (zh) * 2019-09-05 2019-12-13 腾讯科技（深圳）有限公司 点击概率预测、页面内容推荐方法和装置
CN110781389A (zh) * 2019-10-18 2020-02-11 支付宝(杭州)信息技术有限公司 用于为用户生成推荐的方法和系统
CN110837598A (zh) * 2019-11-11 2020-02-25 腾讯科技（深圳）有限公司 信息推荐方法、装置、设备及存储介质
CN111652451A (zh) * 2020-08-06 2020-09-11 腾讯科技（深圳）有限公司 社交关系的获取方法和装置及存储介质
CN112084428A (zh) * 2020-09-17 2020-12-15 辽宁工程技术大学 一种基于耦合网络嵌入及知识图谱的协同过滤推荐方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10846593B2 (en) * 2018-04-27 2020-11-24 Qualcomm Technologies Inc. System and method for siamese instance search tracker with a recurrent neural network
CN110175245A (zh) * 2019-06-05 2019-08-27 腾讯科技（深圳）有限公司 多媒体推荐方法、装置、设备及存储介质
CN111104599B (zh) * 2019-12-23 2023-08-18 北京百度网讯科技有限公司 用于输出信息的方法和装置
CN111400514A (zh) * 2020-03-13 2020-07-10 南京领行科技股份有限公司 一种信息推荐方法、装置和系统
CN112395435A (zh) * 2020-11-17 2021-02-23 华北电力大学扬中智能电气研究中心 一种多媒体资源推荐方法、装置、设备及介质

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130097056A1 (en) * 2011-10-13 2013-04-18 Xerox Corporation Methods and systems for recommending services based on an electronic social media trust model
CN105653535A (zh) * 2014-11-13 2016-06-08 中国科学院沈阳计算技术研究所有限公司 一种媒体资源推荐方法
CN105468723A (zh) * 2015-11-20 2016-04-06 小米科技有限责任公司 信息推荐方法及装置
CN106022938A (zh) * 2016-06-02 2016-10-12 北京奇艺世纪科技有限公司 社交网络的用户社团划分方法和装置
CN108280458A (zh) * 2017-01-05 2018-07-13 腾讯科技（深圳）有限公司 群体关系类型识别方法及装置
CN109544396A (zh) * 2019-01-10 2019-03-29 腾讯科技（深圳）有限公司 账号推荐方法、装置、服务器、终端及存储介质
CN110569437A (zh) * 2019-09-05 2019-12-13 腾讯科技（深圳）有限公司 点击概率预测、页面内容推荐方法和装置
CN110781389A (zh) * 2019-10-18 2020-02-11 支付宝(杭州)信息技术有限公司 用于为用户生成推荐的方法和系统
CN110837598A (zh) * 2019-11-11 2020-02-25 腾讯科技（深圳）有限公司 信息推荐方法、装置、设备及存储介质
CN111652451A (zh) * 2020-08-06 2020-09-11 腾讯科技（深圳）有限公司 社交关系的获取方法和装置及存储介质
CN112084428A (zh) * 2020-09-17 2020-12-15 辽宁工程技术大学 一种基于耦合网络嵌入及知识图谱的协同过滤推荐方法

Also Published As

Publication number Publication date
CN115495598A (zh) 2022-12-20
US20230273951A1 (en) 2023-08-31
JP2024509966A (ja) 2024-03-05
WO2022262603A1 (zh) 2022-12-22

Similar Documents

Publication Publication Date Title
CN111931062B (zh) 一种信息推荐模型的训练方法和相关装置
WO2021233199A1 (zh) 搜索推荐模型的训练方法、搜索结果排序的方法及装置
US10922609B2 (en) Semi-supervised learning via deep label propagation
WO2022016556A1 (zh) 一种神经网络蒸馏方法以及装置
CN116010684A (zh) 物品推荐方法、装置及存储介质
CN114298122B (zh) 数据分类方法、装置、设备、存储介质及计算机程序产品
CN111783903B (zh) 文本处理方法、文本模型的处理方法及装置、计算机设备
CN115631008B (zh) 商品推荐方法、装置、设备及介质
US20220327835A1 (en) Video processing method and apparatus
CN112819024B (zh) 模型处理方法、用户数据处理方法及装置、计算机设备
WO2023050143A1 (zh) 一种推荐模型训练方法及装置
WO2023185925A1 (zh) 一种数据处理方法及相关装置
WO2024041483A1 (zh) 一种推荐方法及相关装置
WO2024002167A1 (zh) 一种操作预测方法及相关装置
CN114201516B (zh) 一种用户画像构建的方法、信息推荐的方法以及相关装置
Cheung et al. Characterizing user connections in social media through user-shared images
CN116310318A (zh) 交互式的图像分割方法、装置、计算机设备和存储介质
CN117711001B (zh) 图像处理方法、装置、设备和介质
CN116205700A (zh) 目标产品的推荐方法、装置、计算机设备和存储介质
WO2024067779A1 (zh) 一种数据处理方法及相关装置
WO2023051678A1 (zh) 一种推荐方法及相关装置
US11068802B2 (en) High-capacity machine learning system
WO2022262561A1 (zh) 多媒体资源的处理方法、装置、设备及存储介质
CN116204709A (zh) 一种数据处理方法及相关装置
CN114529399A (zh) 用户数据处理方法、装置、计算机设备和存储介质

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: 22824041

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 07.05.2024)