CN116932873A

CN116932873A - Video account recommending method, device, equipment, storage medium and program product

Info

Publication number: CN116932873A
Application number: CN202210351819.7A
Authority: CN
Inventors: 罗永盛
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2022-04-02
Filing date: 2022-04-02
Publication date: 2023-10-24

Abstract

The embodiment of the application provides a video account recommending method, device, equipment, storage medium and program product, and relates to the fields of artificial intelligence, cloud technology, map and the like, wherein application scenes comprise but are not limited to scenes such as video account recommending. The method comprises the following steps: acquiring an isopgram, wherein the isopgram is used for representing the association relationship between at least two types of nodes; determining a first feature embedding vector of any account node in the different composition and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the different composition; determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of a plurality of neighbor nodes; recommending the video account corresponding to at least one account node except any account node in the iso-composition based on the second feature embedding vector of any account node.

Description

Video account recommending method, device, equipment, storage medium and program product

Technical Field

The application relates to the technical field of computers, in particular to a video account recommending method, a device, equipment, a storage medium and a program product.

Background

With the development of the mobile internet, short videos gain more and more user preference. Various short video platforms, such as points of view, micro-views, video numbers, etc., are endless. A large number of users shoot short videos and upload the short videos to the short video platforms, and hundreds of millions of short videos are newly generated every day. When users actually use and watch videos, they often pay attention to and love videos published by certain types of video accounts, for example, users who love to watch a variety pay attention to video accounts of a variety, and users who love to play a game pay attention to video accounts of a game. In the prior art, video accounts are recommended to users based on isomorphic diagrams, all nodes in the isomorphic diagrams are the same type of nodes, and the accuracy of recommending the video accounts to the users is not high.

Disclosure of Invention

Aiming at the defects of the existing mode, the application provides a video account recommending method, a device, equipment, a computer readable storage medium and a computer program product, which are used for solving the problem of how to improve the accuracy of recommending video accounts.

In a first aspect, the present application provides a method for recommending a video account, including:

acquiring an isopgram, wherein the isopgram is used for representing the association relationship between at least two types of nodes;

determining a first feature embedding vector of any account node in the different composition and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the different composition; the at least two types of nodes include account nodes;

determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of a plurality of neighbor nodes;

recommending the video account corresponding to at least one account node except any account node in the iso-composition based on the second feature embedding vector of any account node.

In one embodiment, acquiring the iso-composition includes:

acquiring at least two types of nodes, wherein the at least two types of nodes comprise an account node and at least one of a video node, a tag node, a media node, an object characteristic information node, an account category node and an account registration area node;

Determining heterogeneous edges between at least two types of nodes;

an iso-pattern is determined based on the heterogeneous edges between the at least two types of nodes.

In one embodiment, determining a first feature embedding vector for any account node in the heterogram comprises:

determining independent heat coding feature vectors of a plurality of neighbor nodes of any account node in the heterogram, wherein the independent heat coding feature vectors of the plurality of neighbor nodes are used for representing initial features of the plurality of neighbor nodes;

and splicing the independent heat coding feature vectors of each neighbor node in the plurality of neighbor nodes to obtain a first feature embedded vector of any account node.

In one embodiment, determining the second feature embedding vector for any account node based on the first feature embedding vector for any account node and the first feature embedding vectors for the plurality of neighbor nodes comprises:

obtaining a third feature embedding vector of any type of node based on the first feature embedding vector of any type of node, the weight corresponding to the first feature embedding vector of any type of node, the first feature embedding vector of at least one neighbor node of which the node type is the node of the type, and the weight corresponding to the first feature embedding vector of at least one neighbor node;

Obtaining a fourth feature embedding vector of any type of node based on the third feature embedding vector of any type of node and the third feature embedding vector of the neighbor node of any type of node;

and carrying out aggregation treatment on the fourth characteristic embedded vector of any account node and the fourth characteristic embedded vectors of a plurality of neighbor nodes of any account node to obtain a second characteristic embedded vector of any account node.

In one embodiment, any type of node is an account node, and the weight corresponding to the first feature embedding vector of any type of node is determined by:

based on the first feature embedded vector of any account node, the first feature embedded vector of at least one neighbor node and a preset first parameter, determining a weight corresponding to the first feature embedded vector of any account node, wherein the first parameter is used for acquiring an association relationship between any account node and at least one neighbor node.

In one embodiment, any type of node is an account node, and the fourth feature embedding vector of any type of node is obtained based on the third feature embedding vector of any type of node and the third feature embedding vector of the neighbor node of any type of node, including:

The third feature embedded vector of any account node and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes are spliced to obtain a spliced feature embedded vector;

and obtaining a fourth feature embedded vector of any account node based on the spliced feature embedded vector and preset second parameters, wherein the second parameters are used for obtaining the association relationship between any account node and a plurality of neighbor nodes.

In one embodiment, the aggregation processing is performed on the fourth feature embedded vector of any account node and the fourth feature embedded vectors of a plurality of neighboring nodes of any account node to obtain a second feature embedded vector of any account node, including:

obtaining a fifth feature embedded vector of any type of node based on the fourth feature embedded vector of any type of node, the weight corresponding to the fourth feature embedded vector of any type of node, the fourth feature embedded vector of at least one neighbor node and the weight corresponding to the fourth feature embedded vector of at least one neighbor node;

and obtaining a second feature embedded vector of any account node based on the fifth feature embedded vector of any account node and the fifth feature embedded vector of each neighbor node in the plurality of neighbor nodes.

In one embodiment, recommending the video account corresponding to at least one account node except any account node in the heterogram based on the second feature embedding vector of any account node includes:

if the similarity between the second feature embedded vector of any account node and the preset second feature embedded vector of at least one account node except any account node in the heterogeneous graph is larger than a preset similarity threshold, recommending the video account corresponding to the at least one account node.

In a second aspect, the present application provides a recommendation device for a video account, including:

the first processing module is used for acquiring the abnormal composition;

the second processing module is used for determining a first characteristic embedded vector of any account node in the different composition and first characteristic embedded vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the different composition; the at least two types of nodes include account nodes;

the third processing module is used for determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of a plurality of neighbor nodes;

And the fourth processing module is used for recommending the video account corresponding to at least one account node except any account node in the iso-composition based on the second characteristic embedding vector of any account node.

In a third aspect, the present application provides an electronic device, comprising: a processor, a memory, and a bus;

a bus for connecting the processor and the memory;

a memory for storing operation instructions;

and the processor is used for executing the recommendation method of the video account number according to the first aspect of the application by calling the operation instruction.

In a fourth aspect, the present application provides a computer readable storage medium storing a computer program for executing the recommendation method of the video account number of the first aspect of the present application.

In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method for recommending a video account according to the first aspect of the present application.

The technical scheme provided by the embodiment of the application has at least the following beneficial effects:

acquiring an isopgram, wherein the isopgram is used for representing the association relationship between at least two types of nodes; determining a first feature embedding vector of any account node in the different composition and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the different composition; the at least two types of nodes include account nodes; determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of a plurality of neighbor nodes; in this way, through at least two types of nodes in the different composition, the feature representation capability of any account node is enhanced, the feature information of any account node is accurately expressed, namely, the feature information of any account node is accurately expressed by using the second feature embedding vector of any account node, and video accounts corresponding to at least one account node except any account node in the different composition are recommended based on the second feature embedding vector of any account node, so that the accuracy of recommending the video accounts corresponding to at least one account node is improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings that are required to be used in the description of the embodiments of the present application will be briefly described below.

FIG. 1 is a schematic diagram of the operation flow of a GraphSAGE model in the prior art;

fig. 2 is a schematic diagram of a recommendation system architecture of a video account according to an embodiment of the present application;

fig. 3 is a flowchart of a video account recommending method according to an embodiment of the present application;

FIG. 4 is a schematic diagram of an iso-patterning provided by an embodiment of the present application;

FIG. 5 is a schematic structural diagram of an improved graphpage model according to an embodiment of the present application;

fig. 6 is a flowchart of another video account recommending method according to an embodiment of the present application;

fig. 7 is a schematic structural diagram of a video account recommending apparatus according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

Embodiments of the present application are described below with reference to the drawings in the present application. It should be understood that the embodiments described below with reference to the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application, and the technical solutions of the embodiments of the present application are not limited.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and "comprising," when used in this specification, specify the presence of stated features, information, data, steps, operations, elements, and/or components, but do not preclude the presence or addition of other features, information, data, steps, operations, elements, components, and/or groups thereof, all of which may be included in the present specification. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein indicates at least one of the items defined by the term, e.g. "a and/or B" indicates implementation as "a", or as "B", or as "a and B".

It will be appreciated that in the specific embodiment of the present application, recommendation related data relating to video account numbers is required to obtain user permissions or agreements when the above embodiments of the present application are applied to specific products or technologies, and the collection, use and processing of related data is required to comply with relevant laws and regulations and standards of relevant countries and regions.

For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application will be described in further detail with reference to the accompanying drawings.

The embodiment of the application provides a video account recommending method provided by a video account recommending system, and relates to the fields of artificial intelligence, cloud technology, map field and the like. Illustratively, the improved graphSAge model and the like referred to in the embodiments of the present application are artificial neural networks in the field of artificial intelligence. The application scenes of the video account recommending method include but are not limited to scenes such as recommended video accounts.

Artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and other directions.

Natural language processing (Nature Language processing, NLP) is an important direction in the fields of computer science and artificial intelligence. It is studying various theories and methods that enable effective communication between a person and a computer in natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Thus, the research in this field will involve natural language, i.e. language that people use daily, so it has a close relationship with the research in linguistics. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic questions and answers, knowledge graph techniques, and the like.

Machine Learning (ML) is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

The intelligent transportation system (Intelligent Traffic System, ITS), also called intelligent transportation system (Intelligent Transportation System), is a comprehensive transportation system which uses advanced scientific technology (information technology, computer technology, data communication technology, sensor technology, electronic control technology, automatic control theory, operation study, artificial intelligence, etc.) effectively and comprehensively for transportation, service control and vehicle manufacturing, and enhances the connection among vehicles, roads and users, thereby forming a comprehensive transportation system for guaranteeing safety, improving efficiency, improving environment and saving energy.

In order to better understand and describe the schemes of the embodiments of the present application, some technical terms related to the embodiments of the present application are briefly described below.

The figure: graph/Network structure data can naturally express the relation between objects, and is ubiquitous in daily life and work; for example: microblogs and the like form a social network among people; thousands of pages on the internet form a web page link network; the transportation traffic between the national cities forms a logistics network. The graph structure data is taken as a typical non-European data, and analysis of the graph structure data mainly focuses on node classification, link prediction, clustering and the like. For Graph structure data, graph embedding (Graph/Network Embedding) and Graph neural network (Graph Neural Networks, GNN) are two similar areas of research. Graph embedding aims to represent the nodes of the graph as a low-dimensional vector space while preserving the topology and node information of the network so that existing machine learning algorithms can be directly used in subsequent graph analysis tasks. A node in the graph neural network (Graph Neural Network, GNN) can be defined by its features and related nodes, the goal of GNN is to learn a state embedding to represent each node's neighbor information. The graph rolling network GCN is an important branch of the graph neural network, which applies a convolution operation of conventional data (e.g., images) to the graph structure data.

Graph structure data: one data structure describing various complex data objects by the characteristics of and the connective edge relationships between entity nodes can be represented as g= (V, E), where V is a set of nodes (Edges) and E is a set of Edges. Each edge contains two endpoints u and V, where u, V e V. Common graph structure data such as knowledge graph, social network or communication network, etc.

Graphvage model: the graphbag model is a GCN (Graph Convolutional Network, graph roll-up network); the core idea of the graphpage model is to learn not to try to learn the ebedding of all nodes on a graph, but to learn a mapping that yields ebedding for each node. Specifically, instead of training a separate emmbedding vector for each vertex, the graphpage model trains a set aggregator functions of functions that learn how to aggregate feature information from the local neighbors of a vertex (as shown in fig. 1). Each aggregation function aggregates information from different hops or different depths of search for a vertex. During testing or inference, the training system is used to generate ebedding for the completely unseen vertices through the learned aggregation function. The operation flow of the graphpage model is shown in fig. 1, and can be divided into three steps: 1. sampling neighbor vertexes, namely sampling neighbor vertexes of each vertex in the isomorphic graph, wherein the degree of each node is inconsistent, and for calculation efficiency, sampling a fixed number of neighbors for each node; 2. aggregating neighbor vertex characteristic information, and aggregating information contained in neighbor vertices according to an aggregation function; 3. and obtaining vector representation of each vertex in the isomorphic graph by using node information of the aggregated feature prediction graph, and providing the vector representation for downstream tasks.

Single heat coding: one-hot coding (one-hot coding) is typically used to handle features that do not have a size relationship between categories. For example, a feature has a total of four categories (a, B, AB, O); after the single thermal coding is adopted, a certain characteristic can be changed into a 4-dimensional sparse vector, A is (1, 0), B is (0, 1, 0), AB is (0, 1, 0), and O is (0, 1).

ReLU function: the linear rectification function (Linear rectification function), also known as a modified linear unit, is an activation function (activation function) commonly used in artificial neural networks, and generally refers to a nonlinear function represented by a ramp function and its variants.

Verticality: the vertical class refers to the vertical field, the internet industry term, and provides specific services for a defined group; for example, a platform corresponds to a literature user and a platform corresponds to a male sports user.

The scheme provided by the embodiment of the application relates to an artificial intelligence technology, and the technical scheme of the application is described in detail by a specific embodiment. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.

In order to better understand the scheme provided by the embodiment of the present application, the scheme is described below in connection with a specific application scenario.

In an embodiment, fig. 2 shows a schematic architecture diagram of a video account recommendation system to which the embodiment of the present application is applicable, and it can be understood that the video account recommendation method provided by the embodiment of the present application may be applicable, but not limited to, in an application scenario as shown in fig. 2.

In this example, as shown in fig. 2, the architecture of the recommendation system for video accounts in this example may include, but is not limited to, a terminal 10, a server 20, a network 30, and a database 40. Interactions between the terminal 10, the server 20 and the database 40 may be through the network 30. The server 20 acquires an isopgram for characterizing the association relationship between at least two types of nodes; the server 20 determines a first feature embedding vector of any account node in the heterogram and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the heterogram; the at least two types of nodes include account nodes; the server 20 determines a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of the plurality of neighbor nodes; based on the second feature embedded vector of any account node, the server 20 recommends a video account corresponding to at least one account node except any account node in the iso-graph to the terminal 10, and sends the video account corresponding to the at least one account node to the database 40.

It will be appreciated that the above is only an example, and the present embodiment is not limited thereto.

The terminal may be a smart phone (such as an Android phone, an iOS phone, etc.), a phone simulator, a tablet computer, a notebook computer, a digital broadcast receiver, a MID (Mobile Internet Devices, mobile internet device), a PDA (personal digital assistant), etc. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server or a server cluster for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs (Content Delivery Network, content delivery networks), basic cloud computing services such as big data and artificial intelligent platforms, and the like.

Cloud computing (clouding) is a computing model that distributes computing tasks across a large pool of computers, enabling various application systems to acquire computing power, storage space, and information services as needed. The network that provides the resources is referred to as the "cloud". Resources in the cloud are infinitely expandable in the sense of users, and can be acquired at any time, used as needed, expanded at any time and paid for use as needed.

As a basic capability provider of cloud computing, a cloud computing resource pool (cloud platform for short, generally referred to as IaaS (Infrastructure as a Service, infrastructure as a service) platform) is established, in which multiple types of virtual resources are deployed for external clients to select for use.

According to the logic function division, a PaaS (Platform as a Service ) layer can be deployed on an IaaS (Infrastructure as a Service ) layer, and a SaaS (Software as a Service, software as a service) layer can be deployed above the PaaS layer, or the SaaS can be directly deployed on the IaaS. PaaS is a platform on which software runs, such as a database, web container, etc. SaaS is a wide variety of business software such as web portals, sms mass senders, etc. Generally, saaS and PaaS are upper layers relative to IaaS.

The artificial intelligence cloud Service is also commonly referred to as AIaaS (AIas a Service, chinese is "AI as Service"). The service mode of the artificial intelligent platform is the mainstream at present, and particularly, the AIaaS platform can split several common AI services and provide independent or packaged services at the cloud. This service mode is similar to an AI theme mall: all developers can access one or more artificial intelligence services provided by the use platform through an API interface, and partial deep developers can also use an AI framework and AI infrastructure provided by the platform to deploy and operate and maintain self-proprietary cloud artificial intelligence services.

The network may include, but is not limited to: a wired network, a wireless network, wherein the wired network comprises: local area networks, metropolitan area networks, and wide area networks, the wireless network comprising: bluetooth, wi-Fi, and other networks implementing wireless communications. And in particular, the method can be determined based on actual application scene requirements, and is not limited herein.

Referring to fig. 3, fig. 3 is a schematic flow chart of a video account recommendation method provided by the embodiment of the present application, where the method may be performed by any electronic device, for example, a server, and as an optional implementation manner, the method may be performed by the server, and for convenience of description, in the following description of some optional embodiments, a server will be taken as an example of a method execution body. As shown in fig. 3, the method for recommending video account numbers provided by the embodiment of the application includes the following steps:

s201, acquiring an isopgram, wherein the isopgram is used for representing the association relation between at least two types of nodes.

Specifically, the heterogeneous graph may be graph structure data, and the heterogeneous graph may include a plurality of types of nodes, such as an account node, a video node, a tag node, a media node, an object feature information node, an account category node, an account registration area node, and the like. Two nodes in the iso-graph may form one heterogeneous edge. For example, as shown in fig. 4, the heterogram includes an account node a, an account node B, a video node, a tag node and a media node, where the neighbor nodes of the account node a are the account node B, the video node, the tag node and the media node.

S202, determining a first feature embedding vector of any account node in the heterogram and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the heterogram; the at least two types of nodes include account nodes.

Specifically, the neighbor node of any account node is a node directly connected with any account node, and the plurality of neighbor nodes may be multiple types of nodes, for example, the node types of the plurality of neighbor nodes may be account nodes, video nodes, tag nodes, and media nodes. The plurality of neighbor nodes of any account node may be a part of all neighbor nodes of any account node.

S203, determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of the plurality of neighbor nodes.

Specifically, the first feature embedded vector of any account node may be used to characterize the content feature of the account corresponding to any account node, where the content feature of the account may include a video published by the account, an account tag, an account drop class, and the like. And the first feature embedded vector of any account node and the first feature embedded vectors of a plurality of neighbor nodes are subjected to aggregation treatment, so that the feature representation capability of any account node is enhanced, and the feature information of any account node is accurately expressed, namely, the feature information of any account node is accurately expressed by using the second feature embedded vector of any account node.

S204, recommending the video account corresponding to at least one account node except any account node in the iso-composition based on the second feature embedded vector of any account node.

Specifically, in addition to any account node in the heterogram, the heterogram may further include a plurality of account nodes. And determining the similarity between any account node and a plurality of account nodes, sorting the similarity from large to small, taking at least one account node corresponding to the similarity sorted before as the similar account node of any account node, and recommending the video account corresponding to the at least one account node.

In the embodiment of the application, an isophotograph is obtained, and the isophotograph is used for representing the association relationship between at least two types of nodes; determining a first feature embedding vector of any account node in the different composition and first feature embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the different composition; the at least two types of nodes include account nodes; determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of a plurality of neighbor nodes; in this way, through at least two types of nodes in the different composition, the feature representation capability of any account node is enhanced, the feature information of any account node is accurately expressed, namely, the feature information of any account node is accurately expressed by using the second feature embedding vector of any account node, and video accounts corresponding to at least one account node except any account node in the different composition are recommended based on the second feature embedding vector of any account node, so that the accuracy of recommending the video accounts corresponding to at least one account node is improved.

In one embodiment, the acquisition of the iso-composition includes steps A1-A3:

step A1, at least two types of nodes are obtained, wherein the at least two types of nodes comprise an account node and at least one of a video node, a tag node, a media node, an object characteristic information node, an account category node and an account registration area node.

Specifically, the video corresponding to the video node may be video content published by the account, the tag corresponding to the tag node may be tag information of the video, the media corresponding to the media node may be a source of the account, the account type corresponding to the account type node may be a type of the account, and the account registration area corresponding to the account registration area node may be a registration area of the account. Each account has an object that favors a certain direction or layer, for example, the object focusing on a certain account is mostly female that favors cosmetics, and the characteristic information of the type of user is the object characteristic information corresponding to the object characteristic information node.

And step A2, determining heterogeneous edges between at least two types of nodes.

Specifically, heterogeneous edges between multiple types of nodes are determined, and the heterogeneous edges can be account number-account number, account number-video, account number-tag, account number-media and the like.

The heterogeneous edge may be an account number-an account number, for example, two nodes constituting the heterogeneous edge are an account number a and an account number B, respectively; an account number, for example, a user attention account number a and an account number B, may be constructed through the behavior of the user attention account number a and the account number B, wherein the user attention account number a is 100 times, the user attention account number B is 50 times, and a heterogeneous edge is constructed through the account number a and the account number B.

The heterogeneous edge may be an account number-video, for example, two nodes constituting the heterogeneous edge are an account number C and a video D, respectively; an account number-video can be constructed from video content published by account number C.

The heterogeneous edge may be an account number-tag, e.g., two nodes constituting the heterogeneous edge are account number E and tag F, respectively; the statistics and summarization can be carried out from the label information of the video issued by the account E, and the labels with the previous ranks are selected to construct the account-label.

The heterogeneous edge may be account-media, e.g., two nodes constituting the heterogeneous edge are account G and media H, respectively; the account-media may be constructed from the source to which account G belongs (media H).

Step A3, determining an iso-pattern based on the heterogeneous edges between the at least two types of nodes.

Specifically, an iso-pattern is constructed by a plurality of heterogeneous edges.

In one embodiment, determining a first feature embedding vector for any account node in the profile comprises steps B1-B2:

and B1, determining independent heat coding feature vectors of a plurality of neighbor nodes of any account node in the heterogram, wherein the independent heat coding feature vectors of the plurality of neighbor nodes are used for representing initial features of the plurality of neighbor nodes.

Specifically, for example, an account node a is connected with 10 neighbor nodes, and the unique heat coding feature vectors (the 4 neighbor nodes are respectively an account node B, a video node C, a label node D and a media node E) of 4 neighbor nodes with the largest weight can be selected from the 10 neighbor nodes to perform splicing processing, so as to obtain a first feature embedded vector of the account node a; the weight may be determined according to the attention degree of the user, for example, 100 users pay attention to the account B,50 users pay attention to the account F,10 users pay attention to the account G, the attention degree of the account B, the account F and the account G is the account B at the maximum, and the account node B corresponding to the account B is a neighbor node with the maximum weight.

And step B2, splicing the independent heat coding feature vectors of each neighbor node in the plurality of neighbor nodes to obtain a first feature embedded vector of any account node.

Specifically, for example, the neighboring node may be represented as a 100-dimensional one-hot encoded feature vector, the 100-dimensional one-hot encoded feature vector may be represented as [0, …,1] or the like, and the one-hot encoded feature vector of the account node B, the one-hot encoded feature vector of the video node C, the one-hot encoded feature vector of the tag node D, and the one-hot encoded feature vector of the media node E are subjected to a splicing process, to obtain a 400-dimensional feature embedded vector of the account node a, that is, a first feature embedded vector of the account node a.

In one embodiment, determining a second feature embedding vector for any account node based on the first feature embedding vector for any account node and the first feature embedding vectors for the plurality of neighboring nodes, comprises steps C1-C3:

step C1, obtaining a third feature embedding vector of any type of node based on a first feature embedding vector of any type of node, a weight corresponding to the first feature embedding vector of any type of node, a first feature embedding vector of at least one neighbor node of which the node type is the type of node, and a weight corresponding to the first feature embedding vector of at least one neighbor node.

Specifically, as shown in fig. 5, for example, the improved graphpage model includes three layers, a first layer of the three layers is an input layer, the first layer is used for constructing an iso-graph, and the input of the first layer is heterogeneous edges among multiple types of nodes, and the heterogeneous edges can be account number-account number, account number-video, account number-tag, and account number-media; the output of the first layer is an heterogram, where the heterogram includes account nodes (puin fields), video nodes (video fields), tag nodes (tag fields), and media nodes (media fields).

The second of the three layers is a first aggregation layer (first aggregation layer), the third of the three layers is a second aggregation layer (second aggregation layer), and the second and first aggregation layers are identical in structure, wherein the first aggregation layer comprises a Field-level Func module and a Full connection and ReLU module.

First feature embedding vector of any account node and first feature of any account nodeThe weight corresponding to the embedded vector, the first feature embedded vector of at least one neighbor node with the node type being the account node, and the weight corresponding to the first feature embedded vector of at least one neighbor node are calculated by a Field-level Func module in the first aggregation layer to obtain a third feature embedded vector of any account nodeCalculate->As shown in formula (1):

wherein k=1 represents a first aggregation layer, p represents an account node, l represents a sum of numbers between any account node and at least one neighbor node of which the node type is the account node, and a first feature embedding vector of any account node is p _h The first feature embedding vector p of any account node _h The corresponding weight isThe first feature embedded vector of at least one neighbor node with the node type being the account node is p respectively ₁ 、p ₂ …p _l The weights corresponding to the first feature embedding vectors of at least one neighboring node are +.>The number of at least one neighbor node is l-1.

For example, any account node is connected with 100 nodes, that is, the 100 nodes are a plurality of neighbor nodes, the node type of 20 nodes in the 100 nodes is the account node, and l-1 nodes can be selected from the 20 nodes as at least one neighbor node with the node type being the account node.

In one embodiment, the first feature of any video node is embeddedThe vector, the weight corresponding to the first feature embedded vector of any video node, the first feature embedded vector of at least one neighbor node with the node type of the video node, and the weight corresponding to the first feature embedded vector of at least one neighbor node are calculated to obtain a third feature embedded vector of any account node through a Field-level Func module in a first aggregation layerAt least one neighbor node with the node type being a video node is a neighbor node of any video node, k=1 represents a first aggregation layer, and v represents a video node.

Calculating a third feature embedding vector of any video node through the Field-level Func module in the first aggregation layer Calculate->As shown in formula (2):

where k=1 denotes the first aggregation layer, v denotes the label, l denotes the sum of the number between any video node and at least one neighboring node of which the node type is a video node, and the first feature embedding vector of any video node is v _h First feature embedding vector v of any video node _h The corresponding weight isThe first feature embedded vectors of at least one neighbor node with the node type being the video node are v respectively ₁ 、v ₂ …v _l The weights corresponding to the first feature embedding vectors of at least one neighboring node are +.>The number of at least one neighbor node is l-1.

In one embodiment, the first feature embedded vector of any tag node, the weight corresponding to the first feature embedded vector of any tag node, the first feature embedded vector of at least one neighbor node of which the node type is the tag node, and the weight corresponding to the first feature embedded vector of at least one tag node are calculated by the Field-level Func module in the first aggregation layer to obtain the third feature embedded vector of any tag nodeWherein at least one neighbor node of which the node type is a label node is a neighbor node of any label node, k=1 represents the first aggregation layer, and t represents the label node.

Calculating a third feature embedding vector of any label node through the Field-level Func module in the first aggregation layerCalculate->As shown in formula (3):

where k=1 denotes the first aggregation layer, t denotes the label, l denotes the sum of the number between any label node and at least one neighbor node of which the node type is a label node, and the first feature embedding vector of any label node is t _h First feature embedding vector t of any tag node _h The corresponding weight isThe first feature embedded vector of at least one neighbor node with the node type being a label node is t respectively ₁ 、t ₂ …t _l At least one neighbor nodeWeights corresponding to the first feature embedding vectors of (a) are +.> The number of at least one neighbor node is l-1.

In one embodiment, the first feature embedded vector of any media node, the weight corresponding to the first feature embedded vector of any media node, the first feature embedded vector of at least one neighbor node of which the node type is a media node, and the weight corresponding to the first feature embedded vector of at least one media node are calculated by the Field-level Func module in the first aggregation layer to obtain the third feature embedded vector of any media node Wherein at least one neighbor node of which the node type is a media node is a neighbor node of any media node, k=1 represents the first aggregation layer, and m represents the media node.

Calculating a third feature embedding vector of any media node through the Field-level Func module in the first aggregation layerCalculate->As shown in formula (4):

where k=1 denotes the first aggregation layer, m denotes the media, l denotes the sum of the number between any one media node and at least one neighbor node of which the node type is a media node, and the first feature embedding vector of any one media node is m _h First feature embedding vector m of any media node _h Corresponding weightIs thatThe first feature embedded vector of at least one neighbor node with the node type being a media node is m respectively ₁ 、m ₂ …m _l The weights corresponding to the first feature embedding vectors of at least one neighboring node are +.>The number of at least one neighbor node is l-1.

And C2, obtaining a fourth feature embedding vector of the node of any type based on the third feature embedding vector of the node of any type and the third feature embedding vector of the neighbor node of the node of any type.

Specifically, the third feature embedded vector of any account node and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes are calculated through a Full connection and ReLU module in the first aggregation layer to obtain a fourth feature embedded vector y of any account node _k Calculate y _k As shown in formula (5):

wherein, reLU is a ReLU function, W _n Is the second parameter, W _n May be a matrix of parameters that can be learned,and the third characteristic embedding vector of any account node and the third characteristic embedding vector of each neighbor node in the plurality of neighbor nodes are subjected to splicing processing.

For example, the dimension of the learnable parameter matrix is 2x3, which is shown below:

[[-0.82452497-0.82452497-0.82452497]

[-0.01143752-0.01143752-0.01143752]]。

for example, any one ofThe account nodes are connected with 100 nodes, namely the 100 nodes are neighbor nodes of any account node, 20 nodes are selected from the 100 nodes to represent four types of nodes (account nodes, video nodes, label nodes and media nodes), and the 20 nodes comprise 5 account nodes (each account node in the 5 account nodes corresponds to 1) 5 video nodes (1 for each of the 5 video nodes)>) 5 tag nodes (1 for each of the 5 tag nodes)>) And 5 media nodes (1 +.5 for each of the 5 media nodes)>) These 20 nodes and 1 account node (either account node), i.e. 21 nodes are subjected to +.>Finally, obtaining y of any account node _k . Wherein (1) >And->Are all the same dimension, e.g.)>And->May be 100 in dimension.

To 5 video nodesAveraging to obtain data with dimension of 100: [ -0.82452497-0.82452497-0.82452497... -0.01143752-0.01143752]The method comprises the steps of carrying out a first treatment on the surface of the Let 6 account nodes +.>Average values were taken to give data with dimensions of 100 [ -0.2324673-0.72125972-0.2479342. -0.763237512-0.22246112]5 tag nodes +.>Average values were taken to give data with dimensions 100 [0.32479-0.11333452 0.47895341.. 0.11236900350.235462465]5 media nodes +.>Average values were taken to give data with one dimension of 100 [ 0.68924-0.111359746-0.211119397623..-0.454363888-0.11434689 ]]The method comprises the steps of carrying out a first treatment on the surface of the After the 4 pieces of data with 100 dimensions are spliced, data with 400 dimensions are obtained: [ -0.82452497-0.82452497-0.82452497...... -0.01143752-0.01143752-0.2324673-0.72125972-0.2479342... -0.763237512-0.22246112 0.32479-0.11333452 0.47895341...0.1123690035 0.235462465-0.82452497-0.82452497-0.82452497... -0.454363888-0.11434689]。

And C3, carrying out aggregation treatment on the fourth characteristic embedded vector of any account node and the fourth characteristic embedded vectors of a plurality of neighbor nodes of any account node to obtain a second characteristic embedded vector of any account node.

Specifically, the structure of the second aggregate layer is exactly the same as that of the first aggregate layer. Fourth feature embedding vector y of any account node _k Wherein k=1, i.e. y _k ＝y ₁ The method comprises the steps of carrying out a first treatment on the surface of the Embedding a fourth feature of any account node into the vector y ₁ And taking fourth feature embedded vectors of a plurality of neighbor nodes of any account node as input of a second aggregation layer, and performing aggregation processing to obtain output of the second aggregation layer, namely the second feature embedded vectors of any account nodey _k Wherein k=2, i.e. y _k ＝y ₂ 。

Specifically, a weight corresponding to a first feature embedding vector of any account node is calculated, as shown in formula (6):

wherein, the liquid crystal display device comprises a liquid crystal display device,for presetting a first parameter, < > >A learnable parameter matrix that can be an account node; feature embedding vector p _j The corresponding account node is characterized by an embedded vector p _i Neighbor nodes of the corresponding account nodes, n=l, j not equal to i.

In one embodiment, any type of node is an account node, and a fourth feature embedding vector of any type of node is obtained based on a third feature embedding vector of any account node and a third feature embedding vector of a neighbor node of any type of node, including steps D1-D2:

and D1, performing splicing processing on the third feature embedded vector of any account node and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes to obtain a spliced feature embedded vector.

Specifically, any account is savedThe third feature embedded vector of the point and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes are calculated to obtain a fourth feature embedded vector y of any account node through a Full connection and ReLU module in the first aggregation layer _k Calculate y _k As shown in formula (5):

wherein, reLU is a ReLU function, W _n Is the second parameter, W _n May be a matrix of parameters that can be learned,and performing splicing processing on the third feature embedded vector of any account node and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes, so as to obtain a spliced feature embedded vector.

And D2, obtaining a fourth feature embedded vector of any account node based on the spliced feature embedded vector and a preset second parameter, wherein the second parameter is used for obtaining the association relationship between any account node and a plurality of neighbor nodes.

Specifically, based on the feature embedding vector after splicing, the second parameter W is preset _n Obtaining a fourth characteristic embedded vector y of any account node through a ReLU function _k 。

In one embodiment, the aggregation processing is performed on the fourth feature embedded vector of any account node and the fourth feature embedded vectors of a plurality of neighbor nodes of any account node to obtain a second feature embedded vector of any account node, including steps E1-E2:

and E1, obtaining a fifth feature embedded vector of any type of node based on the fourth feature embedded vector of any type of node, the weight corresponding to the fourth feature embedded vector of any type of node, the fourth feature embedded vector of at least one neighbor node and the weight corresponding to the fourth feature embedded vector of at least one neighbor node.

Specifically, based on the same calculation manner of equation (1), equation (2), equation (3), equation (4), and equation (6), where k=2, k=2 represents the second aggregation layer, p _h Embedding a vector y for a fourth feature of any account node ₁ Fourth feature embedding vector p of any account node _h The corresponding weight isThe fourth feature embedded vector of at least one neighbor node with the node type being the account node is p respectively ₁ 、p ₂ …p _l (p ₁ 、p ₂ …p _l The values of which are all the output of the first aggregation layer), the weights corresponding to the fourth feature embedded vectors of at least one neighboring node are respectively +.>The number of at least one neighbor node is l-1, and the fifth feature embedding vector of any account node can be calculated to be +.>

And E2, obtaining a second feature embedded vector of any account node based on the fifth feature embedded vector of any account node and the fifth feature embedded vector of each neighbor node in the plurality of neighbor nodes.

Specifically, based on the same calculation mode of the formula (5), a second feature embedding vector y of any account node is obtained _k Wherein k=2, i.e. y _k ＝y ₂ 。

In one embodiment, in the improved graphpage model training, the third layer output of the improved graphpage model is y _k ＝y ₂ If the preset loss function is smaller than the preset threshold, finishing training of the improved graphpage model; if the predetermined loss function is greater than or equal to the predetermined threshold, the parameters are adjusted, e.gAnd W is _n The improved graphpage model is continuously trained.

The application of the embodiment of the application has at least the following beneficial effects:

based on an improved graphpage model, the characteristic representation capability of any account node is enhanced through at least two types of nodes in the heterograms, so that the characteristic information of any account node is accurately expressed, namely, the characteristic information of any account node is accurately expressed by using a second characteristic embedded vector of any account node, and video accounts corresponding to at least one account node except any account node in the heterograms are recommended based on the second characteristic embedded vector of any account node, so that the accuracy of recommending the video accounts corresponding to at least one account node is improved; the improved graphpage model adopts two aggregation layers (a first aggregation layer and a second aggregation layer), so that the model performance is improved, and the neighbor aggregation capability of the model is improved.

In order to better understand the method provided by the embodiment of the present application, the scheme of the embodiment of the present application is further described below with reference to examples of specific application scenarios.

The method provided by the embodiment of the application can be applied to recommended video account scenes, such as recommended short video account scenes and the like.

In one embodiment, an evaluation set is constructed from a behavioral sequence of accounts of interest to the real user, wherein 5 million users 'account sequences of interest are used as training sets and other 5 million users' account sequences of interest are used as test sets. In the actual test, the first account of interest (the first account of interest corresponds to any account node) of the user is used as a search, the improved graphpage model returns N similar accounts (N similar accounts correspond to at least one account node) with the previous similarity sequence, and the hit ratio (HitRate) between the N accounts and the account sequence of the user actually focused on is calculated. On the effect of the improved graphpage model, the evaluation set tests the DNN (Deep Neural Networks, deep neural network) model which is currently used on the line and the improved graphpage model respectively; compared with the DNN model, the improved graphage model improves the HitRate (top 50) from 9.21% to 12.93%, and the improved graphage model improves the HitRate by about 3.7% points.

Referring to fig. 6, fig. 6 is a flowchart of a video account recommending method according to an embodiment of the present application, where the method may be performed by any electronic device, for example, a server, and as an optional implementation manner, the method may be performed by the server, and for convenience of description, in the following description of some optional embodiments, a description will be given by taking the server as an implementation subject of the method. As shown in fig. 6, the method for recommending video account numbers provided by the embodiment of the application includes the following steps:

s401, acquiring multiple types of nodes, wherein the multiple types of nodes comprise account nodes, video nodes, label nodes and media nodes.

S402, constructing heterogeneous edges among multiple types of nodes.

S403, inputting each heterogeneous edge to an input layer of the improved graphpage model to obtain a heterogram.

S404, determining a first feature embedded vector of an account node in the heterogram and first feature embedded vectors of a plurality of neighbor nodes of the account node.

S405, performing aggregation processing on the first feature embedded vector of the account node and the first feature embedded vectors of a plurality of neighbor nodes through two aggregation layers of the improved Graphsag model to obtain a second feature embedded vector of the account node.

S406, if the similarity between the second feature embedded vector of the account node and the second feature embedded vectors of the account nodes except the account node in the heterogeneous graph is greater than a preset similarity threshold, recommending the short video accounts respectively corresponding to the account nodes.

based on an improved graphpage model, the characteristic representation capability of a certain account node is enhanced through various types of nodes in the different composition, the characteristic information of the account node is accurately expressed, namely, the characteristic information of the account node is accurately expressed by using a second characteristic embedded vector of the account node, short video accounts respectively corresponding to a plurality of account nodes except the account node in the different composition are recommended based on the second characteristic embedded vector of the account node, and therefore accuracy of recommending the short video accounts respectively corresponding to the account nodes is improved.

The embodiment of the application also provides a video account recommending device, and a structural schematic diagram of the video account recommending device is shown in fig. 7, and the video account recommending device 50 includes a first processing module 501, a second processing module 502, a third processing module 503 and a fourth processing module 504.

A first processing module 501, configured to obtain an iso-composition;

the second processing module 502 is configured to determine a first feature embedding vector of any account node in the heterogram, and first feature embedding vectors of a plurality of neighboring nodes of any account node, where the plurality of neighboring nodes are at least two types of nodes in the heterogram; the at least two types of nodes include account nodes;

a third processing module 503, configured to determine a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of the plurality of neighboring nodes;

the fourth processing module 504 is configured to recommend a video account corresponding to at least one account node except any account node in the iso-graph based on the second feature embedding vector of any account node.

In one embodiment, the first processing module 501 is specifically configured to:

determining heterogeneous edges between at least two types of nodes;

In one embodiment, the second processing module 502 is specifically configured to include:

In one embodiment, the third processing module 503 is specifically configured to:

In one embodiment, any type of node is an account node, and the third processing module 503 is specifically configured to:

In one embodiment, the fourth processing module 504 is specifically configured to:

The embodiment of the application also provides an electronic device, a schematic structural diagram of which is shown in fig. 8, and an electronic device 4000 shown in fig. 8 includes: a processor 4001 and a memory 4003. Wherein the processor 4001 is coupled to the memory 4003, such as via a bus 4002. Optionally, the electronic device 4000 may further comprise a transceiver 4004, the transceiver 4004 may be used for data interaction between the electronic device and other electronic devices, such as transmission of data and/or reception of data, etc. It should be noted that, in practical applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 is not limited to the embodiment of the present application.

The processor 4001 may be a CPU (Central Processing Unit ), general purpose processor, DSP (Digital Signal Processor, data signal processor), ASIC (Application Specific Integrated Circuit ), FPGA (Field Programmable Gate Array, field programmable gate array) or other programmable logic device, transistor logic device, hardware components, or any combination thereof. Which may implement or perform the various exemplary logic blocks, modules and circuits described in connection with this disclosure. The processor 4001 may also be a combination that implements computing functionality, e.g., comprising one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

Bus 4002 may include a path to transfer information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect, peripheral component interconnect standard) bus or an EISA (Extended Industry Standard Architecture ) bus, or the like. The bus 4002 can be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in fig. 8, but not only one bus or one type of bus.

Memory 4003 may be, but is not limited to, ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, RAM (Random Access Memory ) or other type of dynamic storage device that can store information and instructions, EEPROM (Electrically Erasable Programmable Read Only Memory ), CD-ROM (Compact Disc Read Only Memory, compact disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store a computer program and that can be Read by a computer.

The memory 4003 is used for storing a computer program for executing an embodiment of the present application, and is controlled to be executed by the processor 4001. The processor 4001 is configured to execute a computer program stored in the memory 4003 to realize the steps shown in the foregoing method embodiment.

Among them, electronic devices include, but are not limited to: a server, etc.

Embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor, implements the steps of the foregoing method embodiments and corresponding content.

The embodiment of the application also provides a computer program product, which comprises a computer program, wherein the computer program can realize the steps and corresponding contents of the embodiment of the method when being executed by a processor.

Based on the same principle as the method provided by the embodiments of the present application, the embodiments of the present application also provide a computer program product or a computer program, which comprises computer instructions stored in a computer-readable storage medium. The computer instructions are read from a computer-readable storage medium by a processor of a computer device, and executed by the processor, cause the computer device to perform the method provided in any of the alternative embodiments of the application described above.

It should be understood that, although various operation steps are indicated by arrows in the flowcharts of the embodiments of the present application, the order in which these steps are implemented is not limited to the order indicated by the arrows. In some implementations of embodiments of the application, the implementation steps in the flowcharts may be performed in other orders as desired, unless explicitly stated herein. Furthermore, some or all of the steps in the flowcharts may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or phases may be performed at the same time, or each of these sub-steps or phases may be performed at different times, respectively. In the case of different execution time, the execution sequence of the sub-steps or stages can be flexibly configured according to the requirement, which is not limited by the embodiment of the present application.

The foregoing is merely an optional implementation manner of some of the implementation scenarios of the present application, and it should be noted that, for those skilled in the art, other similar implementation manners based on the technical ideas of the present application are adopted without departing from the technical ideas of the scheme of the present application, and the implementation manner is also within the protection scope of the embodiments of the present application.

Claims

1. The recommendation method of the video account is characterized by comprising the following steps of:

acquiring an heterogram, wherein the heterogram is used for representing the association relation between at least two types of nodes;

determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of the plurality of neighbor nodes;

recommending the video account corresponding to at least one account node except any account node in the heterogram based on the second feature embedding vector of any account node.

2. The method of claim 1, wherein the acquiring the iso-pattern comprises:

determining a heterogeneous edge between the at least two types of nodes;

3. The method of claim 1, wherein the determining a first feature embedding vector for any account node in the anomaly graph comprises:

determining independent heat coding feature vectors of a plurality of neighbor nodes of any account node in the different composition, wherein the independent heat coding feature vectors of the plurality of neighbor nodes are used for representing initial features of the plurality of neighbor nodes;

and performing splicing processing on the independent heat coding feature vectors of each neighbor node in the plurality of neighbor nodes to obtain a first feature embedded vector of any account node.

4. The method of claim 1, wherein the determining the second feature embedding vector for the any account node based on the first feature embedding vector for the any account node and the first feature embedding vectors for the plurality of neighbor nodes comprises:

Obtaining a third feature embedding vector of the node of any type based on a first feature embedding vector of the node of any type, a weight corresponding to the first feature embedding vector of the node of any type, a first feature embedding vector of at least one neighbor node of the node type which is the node of the type, and a weight corresponding to the first feature embedding vector of the at least one neighbor node;

obtaining a fourth feature embedding vector of the node of any type based on the third feature embedding vector of the node of any type and the third feature embedding vector of the neighbor node of the node of any type;

and carrying out aggregation treatment on the fourth characteristic embedded vector of any account node and the fourth characteristic embedded vectors of a plurality of neighbor nodes of the any account node to obtain a second characteristic embedded vector of the any account node.

5. The method of claim 4, wherein the node of any type is an account node, and the weight corresponding to the first feature embedding vector of the node of any type is determined by:

and determining a weight corresponding to the first feature embedded vector of any account node based on the first feature embedded vector of any account node, the first feature embedded vector of at least one neighbor node and a preset first parameter, wherein the first parameter is used for acquiring an association relationship between any account node and at least one neighbor node.

6. The method according to claim 4, wherein the node of any type is an account node, the obtaining the fourth feature embedding vector of the node of any type based on the third feature embedding vector of the node of any type and the third feature embedding vector of the neighbor node of the node of any type includes:

performing splicing processing on the third feature embedded vector of any account node and the third feature embedded vector of each neighbor node in the plurality of neighbor nodes to obtain a spliced feature embedded vector;

and obtaining a fourth feature embedded vector of any account node based on the spliced feature embedded vector and a preset second parameter, wherein the second parameter is used for obtaining the association relationship between any account node and the plurality of neighbor nodes.

7. The method of claim 4, wherein the aggregating the fourth feature embedding vector of the any account node and the fourth feature embedding vectors of the plurality of neighboring nodes of the any account node to obtain the second feature embedding vector of the any account node comprises:

Obtaining a fifth feature embedded vector of the node of any type based on the fourth feature embedded vector of the node of any type, the weight corresponding to the fourth feature embedded vector of the node of any type, the fourth feature embedded vector of the at least one neighbor node, and the weight corresponding to the fourth feature embedded vector of the at least one neighbor node;

8. The method according to claim 1, wherein the recommending the video account corresponding to the at least one account node except the any account node in the different composition based on the second feature embedding vector of the any account node includes:

if the similarity between the second feature embedded vector of any account node and the preset second feature embedded vector of at least one account node except any account node in the heterogram is larger than a preset similarity threshold, recommending the video account corresponding to the at least one account node.

9. A video account recommendation device, comprising:

the first processing module is used for acquiring the abnormal composition;

the second processing module is used for determining a first characteristic embedding vector of any account node in the abnormal composition and first characteristic embedding vectors of a plurality of neighbor nodes of any account node, wherein the plurality of neighbor nodes are at least two types of nodes in the abnormal composition; the at least two types of nodes include account nodes;

the third processing module is used for determining a second feature embedding vector of any account node based on the first feature embedding vector of any account node and the first feature embedding vectors of the plurality of neighbor nodes;

and the fourth processing module is used for recommending the video account corresponding to at least one account node except any account node in the heterogram based on the second feature embedding vector of any account node.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1-8.

11. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1-8.

12. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method according to any of claims 1-8.