Data processing method and device based on artificial intelligence, electronic equipment, computer readable storage medium and computer program product

Abstract

The invention provides a data processing method and device based on artificial intelligence, electronic equipment, a computer readable storage medium and a computer program product. The method comprises the steps of generating an object relation graph structure based on relation data; performing first embedding processing based on a graph structure similarity mechanism on the object relation graph structure to obtain a first relation feature of the candidate object, and performing second embedding processing based on a random walk mechanism on the object relation graph structure to obtain a second relation feature of the candidate object; based on the first relation feature of the candidate object, the second relation feature of the candidate object and the object feature of the candidate object, generating a team forming probability of the candidate object; and determining a candidate object to be invited of the target object based on the team forming probability of the candidate object, and executing an invitation operation of the corresponding target object for the candidate object to be invited. Through the application, the success rate of inviting the target object to form the team can be improved, so that the man-machine interaction efficiency is improved.

Description

Technical Field

[0001] The present application relates to artificial intelligence technology, and in particular, to a data processing method, device, electronic device, computer-readable storage medium, and computer program product based on artificial intelligence. Background Art

[0002] Artificial Intelligence (AI) is a comprehensive technology in computer science. By studying the design principles and implementation methods of various intelligent machines, machines are enabled to have the functions of perception, reasoning, and decision-making. Artificial intelligence technology is an interdisciplinary subject, involving a wide range of fields, such as natural language processing technology and machine learning / deep learning. With the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0003] In the interaction solutions in the related art, there is an invitation teaming mechanism, that is, at least two objects form a group or team, and interact in a virtual game scene as a group, or interact in a live broadcast scene or an instant messaging scene as a group. In the invitation teaming mechanism, there is an invitation teaming operation for objects. However, the invitation teaming operation in the related art often performs an invitation operation on all friend objects, resulting in a low invitation success rate and ineffective disturbance. Summary of the Invention

[0004] The embodiments of the present application provide a data processing method, device, electronic device, computer-readable storage medium, and computer program product based on artificial intelligence, which can improve the success rate of inviting a target object to team up, thereby improving the human-computer interaction efficiency.

[0005] The technical solution of the embodiments of the present application is implemented as follows:

[0006] The embodiments of the present application provide a data processing method based on artificial intelligence, including:

[0007] Obtain relationship data between multiple objects, and generate an object relationship graph structure based on the relationship data, where the multiple objects include a target object and at least one candidate object;

[0008] Perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain a second relationship feature of each candidate object;

[0009] Obtain the object feature of each candidate object;

[0010] Generate the teaming probability for each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object.

[0011] Based on the teaming probability of each candidate object, determine the candidate objects to be invited for the target object, and perform an invitation operation for the target object on the candidate objects to be invited.

[0012] An embodiment of the present application provides an artificial intelligence-based data processing device, including:

[0013] An acquisition module, configured to acquire relationship data between multiple objects, and generate an object relationship graph structure based on the relationship data, where the multiple objects include a target object and at least one candidate object;

[0014] An embedding module, configured to perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain the second relationship feature of each candidate object;

[0015] An object module, configured to acquire the object feature of each candidate object;

[0016] A probability module, configured to generate the teaming probability for each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object;

[0017] An invitation module, configured to determine the candidate objects to be invited for the target object based on the teaming probability of each candidate object, and perform an invitation operation for the target object on the candidate objects to be invited.

[0018] An embodiment of the present application provides an electronic device, including:

[0019] A memory, configured to store computer-executable instructions;

[0020] A processor, configured to implement the artificial intelligence-based data processing method provided by the embodiment of the present application when executing the computer-executable instructions stored in the memory.

[0021] An embodiment of the present application provides a computer-readable storage medium, storing computer-executable instructions, which are used to cause a processor to implement the artificial intelligence-based data processing method provided by the embodiment of the present application when executed.

[0022] An embodiment of the present application provides a computer program product, including computer-executable instructions, which, when executed by a processor, implement the data processing method based on artificial intelligence provided by the embodiment of the present application.

[0023] The embodiment of the present application has the following beneficial effects:

[0024] Obtain relationship data between multiple objects, and generate an object relationship graph structure based on the relationship data. The multiple objects include a target object and at least one candidate object. Here, it is equivalent to converting the relationship data into the form of a graph structure. Perform a first embedding process on the object relationship graph structure based on the graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on the random walk mechanism to obtain the second relationship feature of each candidate object; obtain the object feature of each candidate object from the relationship data. Here, the features of the candidate object at the relationship level are determined based on two dimensions of the graph structure similarity mechanism and the random walk mechanism respectively. Generate the teaming probability of each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object. Combine the features of the relationship dimension and the object dimension to determine the teaming probability. Based on the teaming probability of each candidate object, determine the candidate objects to be invited for the target object, and perform the invitation operation corresponding to the target object for the candidate objects to be invited. Since the teaming probability takes into account the features of the relationship dimension and the object dimension, the candidate objects to be invited determined based on the teaming probability are more likely to accept the invitation to form a team, which can improve the success rate of the invitation operation and thus improve the human-computer interaction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a schematic structural diagram of a data processing system based on artificial intelligence provided by an embodiment of the present application;

[0026] Figure 2 is a schematic structural diagram of an electronic device provided by an embodiment of the present application;

[0027] Figures 3A to 3C is a schematic flowchart of a data processing method based on artificial intelligence provided by an embodiment of the present application;

[0028] Figure 4 is a schematic interface diagram of a data processing method based on artificial intelligence provided by an embodiment of the present application;

[0029] Figure 5 is a schematic graph structure diagram of a data processing method based on artificial intelligence provided by an embodiment of the present application;

[0030] Figure 6It is an embedding schematic diagram of the data processing method based on artificial intelligence provided by an embodiment of the present application;

[0031] Figure 7 It is a random walk schematic diagram of the data processing method based on artificial intelligence provided by an embodiment of the present application;

[0032] Figure 8 It is an embedding schematic diagram of the data processing method based on artificial intelligence provided by an embodiment of the present application;

[0033] Figure 9 It is a Huffman tree schematic diagram of the data processing method based on artificial intelligence provided by an embodiment of the present application. Detailed implementation manners

[0034] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be construed as limitations on the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0035] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0036] In the following description, the terms "first / second / third" are merely used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second / third" can be interchanged with a specific order or sequence when permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0038] Before further elaborating on the embodiments of the present application, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0039] 1) Neural network: Refers to an artificial neural network. It is an algorithmic mathematical model that mimics the behavioral characteristics of animal neural networks and performs distributed parallel data processing based on artificial intelligence. This network relies on the complexity of the system and adjusts the relationships between a large number of internal nodes to achieve the purpose of processing information.

[0040] 2) Large-scale Information Network Embedding algorithm: It belongs to the graph network representation learning technology. This algorithm defines two similarities on the graph: first-order similarity and second-order similarity. The goal of the algorithm is to preserve these two similarities.

[0041] 3) DeepWalk algorithm: It belongs to the graph network representation learning technology. It realizes node sequences by randomly walking in the network, and then captures the key information between network nodes based on the node sequences.

[0042] In the interaction solutions in the related technologies, there is an invitation teaming mechanism, that is, at least two objects form a group or team, and interact in the virtual game scene as a group, or interact in the live broadcast scene or instant messaging scene as a group. In the invitation teaming mechanism, the invitation teaming operation for objects is involved. Taking the virtual game scene as an example, that is, the player creates a team as the team leader, and then sends the team link to other game players in the friend circle. Other game players join to complete the teaming. This random teaming will cause ineffective disturbance to friends, and the overall ability of the randomly formed team is poor, and it is difficult to win in the interaction in the virtual scene.

[0043] The embodiments of the present application provide an artificial intelligence-based data processing method, device, electronic device, computer-readable storage medium, and computer program product, which can improve the success rate of inviting a team for a target object, thereby improving the human-computer interaction efficiency.

[0044] The following describes the exemplary applications of the electronic device provided in the embodiments of the present application. The electronic device provided in the embodiments of the present application can be implemented as various types of user terminals such as laptop computers, tablet computers, desktop computers, set-top boxes, mobile devices (for example, mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable game devices), in-vehicle terminals, etc., or can be implemented as a server. Below, the exemplary applications when the device is implemented as a terminal or a server will be described.

[0045] Reference Figure 1 , Figure 1 is a schematic diagram of the application mode of the artificial intelligence-based data processing method provided in the embodiments of the present application; for example, the involved server includes: server 200, network 300, and terminal 400. The terminal 400 is connected to the server 200 through the network 300. The network 300 can be a wide area network, a local area network, or a combination of the two.

[0046] The terminal 400 receives a team formation request creation operation of a target object. The terminal 400 generates an object screening request based on the team formation request creation operation and sends the object screening request to the server 200. The server 200 obtains relationship data among multiple objects and generates an object relationship graph structure based on the relationship data, where the multiple objects include the target object and at least one candidate object; performs a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object, and performs a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain a second relationship feature of each candidate object; obtains an object feature of each candidate object from the relationship data; generates a team formation probability for each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object; determines candidate objects to be invited for the target object based on the team formation probability of each candidate object, and the server 200 returns the candidate objects to be invited to the terminal 400, so that the terminal 400 performs an invitation operation for the candidate objects to be invited corresponding to the target object. For example, the terminal 400 sends invitation information corresponding to the target object to the terminal used by the candidate objects to be invited, or the server 200 directly sends invitation information corresponding to the target object to the terminal used by the candidate objects to be invited.

[0047] It should be noted that the data processing method provided in the embodiments of the present application can be applied to a game scenario. For example, in a game scenario, a player can initiate a team formation invitation to other players to form a team so as to interact and win in a virtual scenario as a group. The candidate objects to be invited can be quickly obtained through the data processing method provided in the embodiments of the present application, and then invitation information can be sent to the candidate objects to be invited in a targeted manner, so that the success probability of team formation can be increased, and further the human-computer interaction efficiency can be improved.

[0048] It should be noted that the data processing method provided in the embodiments of the present application can also be applied to a game live broadcast scenario. For example, in a game live broadcast scenario, a user watching the live broadcast can initiate a team formation invitation to other users watching the live broadcast to achieve live broadcast team formation so as to compete for popularity in the game live broadcast scenario. The candidate objects to be invited can be quickly obtained through the data processing method provided in the embodiments of the present application, and then invitation information can be sent to the candidate objects to be invited in a targeted manner, so that the success probability of live broadcast team formation can be increased, and further the human-computer interaction efficiency can be improved.

[0049] In some embodiments, the server 200 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a smart TV, a smart voice interaction device, a smart home appliance, a vehicle terminal, an aircraft, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected by wired or wireless communication means, which are not limited in the embodiments of the present application.

[0050] In some embodiments, the terminal may implement the data processing method based on artificial intelligence provided in the embodiments of the present application by running computer-executable instructions. For example, the computer-executable instructions may be native programs or software modules in an operating system; they may be local (Native) application programs (APPs, Applications), that is, programs that need to be installed in the operating system to run, such as game APPs; they may also be applets, that is, programs that only need to be downloaded to the browser environment to run; they may also be applets that can be embedded in any APP. In short, the above computer programs may be application programs, modules or plug-ins in any form.

[0051] See Figure 2 , Figure 2 is a schematic structural diagram of an electronic device provided in the embodiments of the present application. The electronic device may be a terminal or a server. Figure 2 The terminal 400 shown includes: at least one processor 410, a memory 450, at least one network interface 420, and a user interface 430. Each component in the terminal 400 is coupled together through a bus system 440. It can be understood that the bus system 440 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, in Figure 2 all kinds of buses are labeled as the bus system 440.

[0052] The processor 410 may be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or any conventional processor, etc.

[0053] The user interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons, and controls.

[0054] The memory 450 can be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard disk drives, optical disk drives, etc. The memory 450 optionally includes one or more storage devices that are physically remote from the processor 410.

[0055] The memory 450 includes volatile memory or non-volatile memory, and may also include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), and the volatile memory can be random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0056] In some embodiments, the memory 450 is capable of storing data to support various operations. Examples of such data include programs, modules, and data structures, or subsets or supersets thereof, which are illustrated below.

[0057] The operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and handling hardware-based tasks;

[0058] The network communication module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420. Exemplary network interfaces 420 include: Bluetooth, Wi-Fi (Wireless Fidelity), and USB (Universal Serial Bus), etc.;

[0059] The presentation module 453 is used to enable the presentation of information (such as a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 associated with the user interface 430 (such as a display screen, speaker, etc.);

[0060] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one of one or more input devices 432.

[0061] In some embodiments, the data processing device based on artificial intelligence provided by the embodiments of the present application can be implemented in software. Figure 2 Shown in the memory 450 is the data processing device 455 based on artificial intelligence, which can be software in the form of programs and plugins, etc., including the following software modules: an acquisition module 4551, an embedding module 4552, an object module 4553, a probability module 4554, and an invitation module 4555. These modules are logical, so they can be combined arbitrarily or further split according to the functions implemented. The functions of each module will be described below.

[0062] Combined with the exemplary applications and implementations of the terminal provided by the embodiments of the present application, the data processing method based on artificial intelligence provided by the embodiments of the present application will be described.

[0063] Next, the data processing method based on artificial intelligence provided by the embodiments of the present application will be described. As before, the electronic device implementing the data processing method based on artificial intelligence of the embodiments of the present application can be a terminal or a server. Taking the server as an example for description. Therefore, the execution subjects of each step will not be repeated hereinafter.

[0064] The data processing method based on artificial intelligence provided by the embodiments of the present application can be applied to the virtual scene of a game to screen out players with the willingness to form a team and reduce the disturbance to players with low willingness to form a team. The game will be used as an application scenario for description below.

[0065] See Figure 3A , Figure 3A which is a schematic flowchart of the data processing method based on artificial intelligence provided by the embodiments of the present application, and will be described in combination with Figure 3A the steps shown.

[0066] In step 101, relationship data between multiple objects is acquired, and an object relationship graph structure is generated based on the relationship data, where the multiple objects include a target object and at least one candidate object.

[0067] As an example, the target object here is an object with a team formation requirement. For example, player A hopes to form a team as the team leader, and in response to the team formation request triggered by player A, player A is taken as the target object. The candidate object here can be an object having an association relationship with the target object. The association relationship here has different dimensions, and the association relationship can be a direct association relationship or an indirect association relationship. For example, there is a conversation relationship between the target object and object A, and there is a conversation relationship between object A and object B. Although there is no direct association relationship between the target object and object B, it can be considered that there is an indirect association relationship between the target object and object B.

[0068] In some embodiments, the obtaining of the relationship data among multiple objects in step 101 can be achieved through the following technical solutions: performing at least one of the following processes: obtaining the social relationship data among the multiple objects; obtaining the conversation relationship data among the multiple objects; obtaining the role relationship data among the multiple objects; obtaining the teaming relationship data among the multiple objects; generating the relationship data based on at least one of the social relationship data, the conversation relationship data, the role relationship data, and the teaming relationship data. Through the embodiments of the present application, relationship data can be obtained from multiple dimensions, thereby improving the representation ability of subsequent relationship features.

[0069] As an example, the social relationship here refers to the friendship relationship. For example, if object A and object B are friends with each other in the virtual scene, then the friendship relationship between object A and object B is the social relationship data; the conversation relationship refers to that there has been a conversation between two objects. For example, if object A and object B have had a conversation in the virtual scene, then the conversation relationship between object A and object B is the conversation relationship data; the role relationship here refers to the relationship between the virtual roles of objects in the game. For example, the role relationship here can be the sworn brotherhood relationship, the master-apprentice relationship, the fellow disciple relationship, etc. For example, if object A and object B were or are currently in a sworn brotherhood relationship in the virtual scene, then the role relationship between object A and object B is the role relationship data; the teaming relationship refers to having been or currently being in the same team. For example, if object A and object B were or are currently in the same team in the virtual scene, then the teaming relationship between object A and object B is the teaming relationship data.

[0070] In some embodiments, the relationship data includes relationship data of at least one relationship dimension; the object relationship graph structure includes the object relationship graph structure of at least one of the relationship dimensions. The generating of the object relationship graph structure based on the relationship data in step 101 can be achieved through the following technical solutions: for the relationship data of each relationship dimension, performing the following processes; generating a node corresponding to each object based on each object; when the relationship data of the relationship dimension indicates that there is a relationship corresponding to the relationship dimension between a first object and a second object, generating an associated edge between the nodes respectively corresponding to the first object and the second object, where the first object and the second object are from the multiple objects; constructing the object relationship graph structure corresponding to the relationship dimension based on the nodes and the associated edges. Through the embodiments of the present application, the object relationship graph structure can be used to express the relationship data, thereby improving the subsequent data processing efficiency.

[0071] Here, for the relationship data of each relationship dimension, an object relationship graph structure will be generated.

[0072] Object relationship graph structure for social relationships (friend network): Players with friend relationships in the game are grouped into an object relationship graph structure corresponding to the friend relationship. Players with friend relationships in the game are selected for a period of time to construct an object relationship graph structure for the friend relationship. In the object relationship graph structure, the nodes are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a friend relationship between the players corresponding to the two connected nodes.

[0073] Object relationship graph structure for conversation relationships (conversation network): Players with conversation behaviors in the game are grouped into an object relationship graph structure corresponding to the conversation relationship. Players with conversation behaviors in the game are selected for a period of time to construct an object relationship graph structure for the conversation relationship. In the object relationship graph structure, the nodes are player account identifiers, and the associated edges in the object relationship graph structure indicate that there is a conversation relationship between the players corresponding to the two connected nodes. The weight of the associated edge can be positively correlated with the number of conversations.

[0074] Object relationship graph structure for sworn brotherhood relationships (sworn brotherhood network): Players with sworn brotherhood relationships in the game are grouped into an object relationship graph structure corresponding to the sworn brotherhood relationship. Players with sworn brotherhood relationships in the game are selected for a period of time to construct an object relationship graph structure for the corresponding sworn brotherhood relationship. In the object relationship graph structure, the nodes are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a sworn brotherhood relationship between the players corresponding to the two connected nodes.

[0075] Object relationship graph structure for master - apprentice relationships (master - apprentice network): Players with master - apprentice relationships in the game are grouped into an object relationship graph structure corresponding to the master - apprentice relationship. Players with master - apprentice relationships in the game are selected for a period of time to construct an object relationship graph structure for the corresponding master - apprentice relationship. In the object relationship graph structure, the nodes are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a master - apprentice relationship between the players corresponding to the two connected nodes.

[0076] Object relationship graph structure for team relationships (clan network): Players with clan relationships in the game are grouped into an object relationship graph structure corresponding to the clan relationship. Players with clan relationships in the game are selected for a period of time to construct an object relationship graph structure for the corresponding clan relationship. In the object relationship graph structure, the nodes are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a clan relationship between the players corresponding to the two connected nodes.

[0077] In step 102, a first embedding process based on a graph structure similarity mechanism is performed on the object relationship graph structure to obtain the first relationship features of each candidate object, and a second embedding process based on a random walk mechanism is performed on the object relationship graph structure to obtain the second relationship features of each candidate object.

[0078] As an example, here, embedding processing is respectively performed on the object relationship graph structure for two different mechanisms to obtain relationship features corresponding to the two mechanisms (the first relationship feature and the second relationship feature). Since the graph structure similarity mechanism constrains the minimization of the similarity before and after the graph structure embedding processing, the first relationship feature of each node is learned so that the graph structure before and after the embedding processing does not change. Since the random walk mechanism is based on the node sequence for learning, the second relationship feature of each node learned can represent the relationship between nodes. It is equivalent to that the former focuses on the relationship between each node as a whole, and the latter focuses on the relationship between each node locally.

[0079] In step 103, the object features of each of the candidate objects are obtained.

[0080] As an example, the object features here can be obtained based on the basic data of the object. The basic data here includes the account of the object, the battle achievements of the object, the interaction data of the object, the virtual resource transfer data of the object, etc. The virtual resource transfer data can be the usage data of the virtual resource, the acquisition data of the virtual resource, etc.

[0081] In step 104, based on the first relationship feature of each of the candidate objects, the second relationship feature of each of the candidate objects, and the object feature of each of the candidate objects, the teaming probability of each of the candidate objects is generated.

[0082] In some embodiments, the generating the teaming probability of each of the candidate objects based on the first relationship feature of each of the candidate objects, the second relationship feature of each of the candidate objects, and the object feature of each of the candidate objects in step 104 can be implemented by the following technical solution: the following processing is performed for each of the candidate objects: performing a mapping process on the first relationship feature of the candidate object and the second relationship feature of the candidate object to obtain the comprehensive relationship feature of the candidate object; performing a pooling fusion process on the comprehensive relationship feature of the candidate object and the object feature of the candidate object to obtain the pooling fusion feature of the candidate object; performing a multi-layer perception process on the candidate object to obtain the teaming probability of the candidate object. Through the embodiments of the present application, the relationship dimension features and the features of the object itself can be combined to predict the teaming probability of the candidate object, improving the prediction accuracy, and the non-linearity degree in the prediction process can be improved through the multi-layer perception process and the pooling fusion process, and the prediction accuracy can also be improved.

[0083] As an example, the first relationship feature and the second relationship feature are mapped here, and the two types of relationship features can be spliced or fused. During the fusion process, weights can be set for different types of relationship features. For example, the node distribution uniformity of the object relationship graph structure can be obtained. The weight of the first relationship feature is negatively correlated with the node distribution uniformity, and the weight of the second relationship feature is positively correlated with the node distribution uniformity. It is also possible to directly select the first relationship feature and the second relationship feature, that is, the mapping result is only the first relationship feature or only the second relationship feature. The basis for selection can also be the node distribution uniformity of the object relationship graph structure. When the node distribution uniformity of the object relationship graph structure is less than the first uniformity threshold, the weight of the second relationship feature is zero, that is, the mapping result is the first relationship feature. When the node distribution uniformity of the object relationship graph structure is greater than the second uniformity threshold, the weight of the first relationship feature is zero, that is, the mapping result is the second relationship feature, and the first uniformity here is less than the second uniformity.

[0084] As an example, the comprehensive relationship features of the candidate objects and the object features of the candidate objects are pooled and fused, which is equivalent to summing multiple input vectors. The multi-layer perception processing is actually a nonlinear calculation of the pooled fusion features, which increases the nonlinearity of the neural network perceptron, see formula (1):

[0085] y=relu (relu (XW1+ bias1)W2+ bias2) (1);

[0086] Among them, X is the pooled fusion feature, W1, bias1, W2 and bias2 are trainable parameters, and y is the team probability.

[0087] In step 105, based on the team formation probability of each candidate object, candidate objects to be invited of the target object are determined, and an invitation operation corresponding to the target object is performed on the candidate objects to be invited.

[0088] As an example, here the candidates are sorted in descending order based on the teaming probability, and the top-ranked candidates are selected as candidates to be invited, or the candidates whose teaming probability is greater than the teaming probability threshold are selected as candidates to be invited.

[0089] In some embodiments, the object relationship graph structure includes an object relationship graph structure corresponding to each of the relationship dimensions; see Figure 3B In step 102, the object relationship graph structure is subjected to a first embedding process based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object. Figure 3B Steps 1021 to 1023 are implemented as shown.

[0090] In step 1021, for the object relationship graph structure of each of the said relationship dimensions, a first embedding process based on a graph structure similarity mechanism is performed on the object relationship graph structure of the relationship dimension, to obtain the first relationship feature corresponding to each candidate object for the said relationship dimension.

[0091] As an example, here it is equivalent to performing a first embedding process based on a graph structure similarity mechanism on the object relationship graph structure of each relationship dimension respectively.

[0092] In some embodiments, the object relationship graph structure of the relationship dimension includes nodes corresponding to each object and association edges between multiple said nodes, and the association edges represent that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; in step 1021, performing a first embedding process based on a graph structure similarity mechanism on the object relationship graph structure of the relationship dimension to obtain the first relationship feature corresponding to each candidate object for the said relationship dimension can be achieved through the following technical solution: obtaining the first probability distribution of each of the association edges in the object relationship graph structure of the relationship dimension, and performing vector initialization processing on each of the nodes in the object relationship graph structure of the relationship dimension to obtain the first embedding feature of each of the nodes; determining the second probability distribution of each of the association edges based on the first embedding feature of each of the nodes; determining a first similarity loss based on the first probability distribution of each of the association edges and the second probability distribution of each of the association edges; updating the first embedding feature of each of the nodes based on the first similarity loss to obtain the second embedding feature of each of the nodes; for each of the candidate objects, taking the second embedding feature of the node corresponding to the candidate object as the first relationship feature corresponding to the candidate object for the said relationship dimension. Through the embodiments of the present application, the graph structure relationship before and after embedding can be constrained as a whole to remain unchanged, so that the learned first relationship feature can effectively represent the relationship of the nodes and improve the subsequent prediction accuracy.

[0093] As an example, the solutions for determining the first probability distribution, the second probability distribution, and the first similarity loss will be explained in detail below. After obtaining the first similarity loss, the first embedding feature of the node is updated with the goal of minimizing the first similarity loss to obtain the second embedding feature of the node, and the second embedding feature here is the embedding feature of the node when the first similarity loss converges to the minimum value.

[0094] In some embodiments, the first probability distribution of each of the associated edges in the object relationship graph structure of the relationship dimension can be implemented through the following technical solution: Obtain the weight of each of the associated edges in the object relationship graph structure of the relationship dimension; perform a summation process on the weights of all the associated edges in the object relationship graph structure of the relationship dimension to obtain a first summation result; use the ratio of the weight of each associated edge to the first summation result as the first probability distribution of each associated edge. By using the weights of the associated edges in the embodiments of the present application to determine the first probability distribution, the first probability distribution can represent the proportion degree of the weights of each associated edge, and further, the first probability distribution can fit the joint distribution of the two nodes corresponding to the associated edge in the entire graph structure.

[0095] As an example, the line algorithm based on first-order similarity mainly fits the joint distribution (first probability distribution) between nodes. For any edge (v i ,v j ) ∈ E, its distribution is defined by formula (2):

[0096]

[0097] where W is the sum of the weights of all the associated edges in the graph structure (the first summation result), w ij is the weight corresponding to the associated edge (v i ,v j ), is the joint distribution (first probability distribution) on the associated edge (v i ,v j ), and v i ,v j are the two nodes connected by the associated edge.

[0098] In some embodiments, the second probability distribution of each of the associated edges can be determined based on the first embedding feature of each of the nodes through the following technical solution: For each of the associated edges, perform the following processing: perform a dot product process on the first embedding feature of the first node and the first embedding feature of the second node to obtain a first dot product result, where the first node and the second node are the two nodes connected by the associated edge; perform a probability function mapping process on the first dot product result to obtain the second probability distribution of the associated edge. By using the inner product of the embedding features in the embodiments of the present application to determine the second probability distribution, the second probability distribution can further fit the joint distribution of the two nodes corresponding to the associated edge in the entire graph structure.

[0099] As an example, the probability function mapping process can refer to formula (3):

[0100]

[0101] Among them, u i , u j are the vectorized representations of nodes v i and nodes v j after embedding processing respectively. p1(v i , v j ) is the distribution (second probability distribution) on (v i , v j ) ∈ E after embedding processing.

[0102] In some embodiments, the above-mentioned determination of the first similarity loss based on the first probability distribution of each associated edge and the second probability distribution of each associated edge can be implemented through the following technical solutions: For each of the associated edges, perform the following processing: Take the logarithm of the second probability distribution to obtain a first logarithmic result; Multiply the first logarithmic result by the first probability distribution to obtain a first multiplication result; Sum the first multiplication results corresponding to multiple associated edges to obtain a second summation result, and take the opposite of the first summation result as the first similarity loss. Through the embodiments of the present application, the first-order similarity can be maintained by constraining the above two probability distributions to be as close as possible, thereby improving the representation ability of the learned first relationship feature.

[0103] As an example, to maintain the first-order similarity and make the above two probability distributions as close as possible, the objective function (first similarity loss) is defined as the divergence of the above two probability distributions, as shown in formula (4):

[0104]

[0105] Among them, O is the divergence of the above two distributions, E is the set of associated edges, (v i , v j ) ∈ E is any associated edge, is the distribution (second probability distribution) on any edge (v i , v j ) ∈ E, p1(v i , v j ) is the distribution (first probability distribution) on (v i , v j ) ∈ E after embedding processing.

[0106] In step 1022, when the number of relationship dimensions is multiple, for each candidate object, fuse the first relationship features of the corresponding multiple relationship dimensions of the candidate object to obtain the first relationship feature of the candidate object.

[0107] In step 1023, when the number of the relationship dimensions is one, for each of the candidate objects, the first relationship feature corresponding to the relationship dimension of the candidate object is used as the first relationship feature of the candidate object.

[0108] As an example, for the case of multiple relationship dimensions, there will be multiple object relationship networks. For an object relationship network with one relationship dimension, a first relationship feature of one relationship dimension can be obtained. For an object relationship network with multiple relationship dimensions, first relationship features corresponding to multiple relationship dimensions can be obtained. Here, a weighted sum processing is performed on the first relationship features corresponding to multiple relationship dimensions to obtain the first relationship feature of the candidate object. The weights for the weighted sum here refer to the weights of each relationship dimension, which can be pre-configured or configured according to the importance degree of each relationship dimension, that is, the weight is positively correlated with the importance degree of the relationship dimension. For the case of one relationship dimension, there is only one object relationship network, and the first relationship feature obtained based on the one object relationship network is directly used as the first relationship feature of the candidate object.

[0109] Through the embodiments of the present application, the first relationship features of multiple object relationship graph structures can be fused, so that the first relationship feature can represent the relationship information of each relationship dimension and improve the representation ability of the first relationship feature.

[0110] In some embodiments, the object relationship graph structure includes an object relationship graph structure corresponding to each of the relationship dimensions; refer to Figure 3C , in step 102, a second embedding process based on a random walk mechanism is performed on the object relationship graph structure to obtain a second relationship feature of each of the candidate objects, which can be implemented through Figure 3C the steps 1024 to 1026 shown.

[0111] In step 1024, for the object relationship graph structure of each of the relationship dimensions, a second embedding process based on a random walk mechanism is performed on the object relationship graph structure of the relationship dimension to obtain a second relationship feature corresponding to each of the candidate objects for the relationship dimension.

[0112] As an example, here it is equivalent to performing a second embedding process based on a random walk mechanism on the object relationship graph structure of each relationship dimension respectively.

[0113] In some embodiments, the object relationship graph structure of the relationship dimension includes nodes corresponding to each object and association edges between multiple nodes, and the association edges represent that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; in step 1024, a second embedding process based on a random walk mechanism is performed on the object relationship graph structure of the relationship dimension to obtain a second relationship feature corresponding to the relationship dimension of each candidate object, which can be implemented by the following technical solution: perform a random walk process based on the association edges on the object relationship graph structure of the relationship dimension to obtain a plurality of node sequences; sequentially perform the following processing for each node sequence: perform an initialization process on each node in the node sequence to obtain a third embedding feature of each node in the node sequence, and based on the node sequence, perform an update process on the third embedding feature of each node in the node sequence to obtain a fourth embedding feature of each node in the node sequence; perform a sampling process based on the fourth embedding feature of each node to obtain a sampled embedding feature of each node; for each candidate object, use the sampled embedding feature of the node corresponding to the candidate object as the second relationship feature corresponding to the relationship dimension of the candidate object. Through the embodiments of the present application, the second relationship feature can be learned by means of natural language processing, so that the relationship between multiple nodes in the graph structure can be concerned, and the calculation complexity can be reduced and the data processing efficiency can be improved by means of sampling.

[0114] As an example, see Figure 7 , and use the random walk algorithm to extract node sequences from the object relationship graph structure. The so-called random walk is to continuously randomly select a walk path on the object relationship graph structure, and finally form a path that runs through the network. Starting from a specific node, each step of the walk randomly selects an edge connected to the current node and moves to the next node along the selected edge, and this process is repeated continuously. Figure 7 The bold trajectory 701 shown is the path obtained by random walk.

[0115] In some embodiments, the above-mentioned initialization process for each node in the node sequence to obtain a third embedding feature of each node in the node sequence can be implemented by the following technical solution: perform the following processing for each node: when the node has not been subjected to the initialization process, perform a random initialization process on the node to obtain a third embedding feature of the node; when the node has been subjected to the initialization process, use the fourth embedding feature obtained by the node's most recent update as the third embedding feature of the node. Through the embodiments of the present application, the embedding feature obtained by the node's last update can be retained during each update, so that the information learned each time is not lost, and the feature learning effect of the node is improved.

[0116] As an example, since multiple node sequences can be obtained, assume that there are node sequence A and node sequence B. Node sequence A includes nodes 1, 2, and 3, and node sequence B includes nodes 1, 4, and 5. First, execute the initialization scheme and the update scheme based on node sequence A. At this time, since node 1 in node sequence A has not participated in training, the initialization of node 1 is randomly initialized to obtain the third embedding feature of node 1, and then the third embedding feature of node 1 is updated to obtain the fourth embedding feature of node 1. After sequentially updating nodes 2 and 3, continue to execute the initialization scheme and the update scheme based on node sequence B. At this time, since node 1 in node sequence B has already participated in training once, the initialization of node 1 is actually using the fourth embedding feature updated during the learning process based on node sequence A last time as the initialization feature before this round of update of node 1, that is, the current third embedding feature of node 1.

[0117] In some embodiments, the above-mentioned process of updating the third embedding feature of each node in the node sequence to obtain the fourth embedding feature of each node in the node sequence can be implemented by the following technical solution: For each node, perform the following processing: Perform context node prediction processing on the node sequence based on the third embedding feature of the node to obtain a node prediction result; Update the third embedding feature of the node based on the error between the real node represented by the node sequence and the node prediction result to obtain the fourth embedding feature of the node. By using the natural language processing method in the embodiments of the present application to learn the second relationship feature, the relationship between multiple nodes in the graph structure can be concerned.

[0118] As an example, see Figure 8 , the process of updating the third embedding feature to obtain the fourth embedding feature borrows the idea of natural language processing. The generated node sequence is regarded as a sentence composed of words, and all sequences can be regarded as a corpus. Finally, use natural language processing tools to represent each node as a vector of dimension d (the third embedding feature). Since the node itself cannot be calculated, a mapping function φ is introduced. Here, the mapping function φ is actually a matrix (the fourth embedding feature), and there are a total of V * d parameters. Here, V is the number of nodes. Figure 8 The sequence used for learning in j (node 1) is mapped to the vector representation φ(v j ) of node v j (that is, the shaded part of the mapping function φ in Figure 8 ). See Figure 9 , through Figure 9The shown Huffman maximum likelihood function calculates the conditional probability, and thus uses the calculated conditional probability to optimize and learn the mapping function φ so as to learn a vectorized representation that can characterize the node. The Huffman maximum likelihood function is used to convert the context problem in natural language processing into the probability of walking from node 1 to leaf nodes (node 5 and node 3) on the path (each leaf node represents a node, forming a complete binary tree).

[0119] As an example, the non-repeated nodes in the node sequence set are used to form a dictionary of size V, which is convenient to express each node in the form of embedding features. A path obtained by random walk is regarded as a sequence. For example, the length of the node sequence V1, V2, V3, V4, V5, V6, V7, V8, V9 is 9. The context words can be predicted using the central word to maximize the context contribution. The window length is set to 4 (the maximum distance between the current word and the predicted word in the sentence) and the size is 4. The central node V5 is used to predict its first 4 nodes and the last 4 nodes (referred to as the context). xi is the third embedding feature of the node. The optimization objective P(x1,x2,x3,x4,x6,x7,x8,x9|x5) is maximized. Specifically, the one-hot encoding of the input node V5 is x5 = [0,0,0,0,1,0,0,0,0,0]. x5 is mapped through the mapping function φ to obtain the third embedding feature, and then passes through 8 prediction mappings to obtain 8 vectors of dimension 1x10. The 8 vectors of dimension 1x10 are the expression vectors of the predicted context. The 8 vectors of dimension 1x10 are respectively normalized through the maximum likelihood function to obtain probability values of dimension 1X10. For example, the first vector corresponds to x1, and the probability value of indexing node V1 should be the largest; the second vector corresponds to x2, and the probability value of indexing node V2 should be the largest, and so on. If the vertex represented by the vector is not the expected prediction, the mapping matrix will be updated by backpropagation. After training, the optimal solution is obtained. At this time, the one-hot encoding input is mapped through the mapping function φ to obtain a 1xd vector, which is the fourth embedding feature of the current node (node V5).

[0120] In some embodiments, the above-mentioned sampling process based on the fourth embedding features of each node to obtain the sampled embedding features of each node can be implemented through the following technical solutions: For each node in the object relationship graph structure, the following processes are performed: Obtain candidate nodes adjacent to the node; Obtain a third node and a fourth node having a connection relationship from the candidate nodes, and obtain a third node and a fifth node not having the connection relationship from the candidate nodes; Determine the first transposed result of the fourth embedding feature of the fourth node and the second transposed result of the fourth embedding feature of the fifth node; Determine the first inner product of the fourth embedding feature of the third node and the first transposed result, and the second inner product of the fourth embedding feature of the third node and the second transposed result; Obtain the embedding feature that is positively correlated with the first inner product and negatively correlated with the second inner product as the sampled embedding feature of the node. By sampling nodes through the embodiments of the present application, the computational complexity can be reduced and the data processing efficiency can be improved.

[0121] As an example, after obtaining the fourth embedding features of each node, the fourth embedding features are input into a sampling function to obtain the sampled embedding features. Here, for node A, all candidate nodes of this node A (for example, 10) will be obtained. The candidate nodes can be the first-order adjacent nodes of node A. Positive and negative samples are sampled from the 10 candidate nodes, and the embedding feature of node A is calculated through formula (5):

[0122]

[0123] where, v i , v j is a pair of positive samples (having an associated edge), v i , v n is a pair of negative samples (not having an associated edge), P n (v)~d v , K is the number of negative sampling times, v i is the third node, v j is the fourth node, is the fourth embedding feature of the third node, is the first transposed result of the fourth embedding feature of the fourth node, v n is the fifth node, is the second transposed result of the fourth embedding feature of the fifth node. v n ~P n (v) represents that the nodes are all in the vector space.

[0124] As an example, the number of the fifth nodes here can be multiple, that is, multiple fifth nodes that have no connection relationship with the third node are obtained. Then, multiple second inner products can be obtained here. Finally, the embedding feature that is positively correlated with the first inner product and negatively correlated with the fusion result of multiple second contents is used as the sampling embedding feature of the node as the sampling embedding feature.

[0125] In step 1025, when the number of the relationship dimensions is multiple, for each of the candidate objects, the second relationship features corresponding to the multiple relationship dimensions of the candidate object are fused to obtain the second relationship feature of the candidate object.

[0126] In step 1026, when the number of the relationship dimensions is one, for each of the candidate objects, the second relationship feature corresponding to the relationship dimension of the candidate object is used as the second relationship feature of the candidate object.

[0127] As an example, for the case of multiple relationship dimensions, there will be multiple object relationship networks. For an object relationship network with one relationship dimension, a second relationship feature of one relationship dimension can be obtained. For an object relationship network with multiple relationship dimensions, second relationship features corresponding to multiple relationship dimensions can be obtained. Here, the second relationship features corresponding to multiple relationship dimensions are weighted and summed to obtain the second relationship feature of the candidate object. The weights for the weighted sum here refer to the weights of each relationship dimension, which can be pre-configured or configured according to the importance degree of each relationship dimension, that is, the weights are positively correlated with the importance degree of the relationship dimensions. For the case of one relationship dimension, there is only one object relationship network, and the second relationship feature obtained based on one object relationship network is directly used as the second relationship feature of the candidate object.

[0128] Through the embodiments of the present application, the second relationship features of multiple object relationship graph structures can be fused, so that the second relationship features can represent the relationship information of each relationship dimension and improve the representation ability of the second relationship features.

[0129] By obtaining relationship data among multiple objects through embodiments of the present application and generating an object relationship graph structure based on the relationship data, the multiple objects include a target object and at least one candidate object. Here, it is equivalent to converting the relationship data into a graph structure form. Perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain the second relationship feature of each candidate object; obtain the object feature of each candidate object from the relationship data. Here, the features of the candidate objects at the relationship level are determined from two dimensions of the graph structure similarity mechanism and the random walk mechanism respectively. Based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object, generate the teaming probability of each candidate object. Combine the features of the relationship dimension and the object dimension to determine the teaming probability. Based on the teaming probability of each candidate object, determine the candidate objects to be invited for the target object, and perform an invitation operation corresponding to the target object for the candidate objects to be invited. Since the teaming probability takes into account the features of both the relationship dimension and the object dimension, the candidate objects to be invited determined based on the teaming probability are more likely to accept the invitation to form a team, which can improve the success rate of the invitation operation and thus improve the human-computer interaction efficiency.

[0130] Next, an exemplary application of the embodiments of the present application in an actual application scenario will be described.

[0131] The terminal receives a teaming request creation operation of the target object, generates an object screening request based on the teaming request creation operation, and sends the object screening request to the server. The server obtains the relationship data among multiple objects and generates an object relationship graph structure based on the relationship data. Among them, the multiple objects include the target object and at least one candidate object; perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain the second relationship feature of each candidate object; obtain the object feature of each candidate object from the relationship data; generate the teaming probability of each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object; determine the candidate objects to be invited for the target object based on the teaming probability of each candidate object, and the server returns the candidate objects to be invited to the terminal so that the terminal performs an invitation operation corresponding to the target object for the candidate objects to be invited. For example, the terminal sends the invitation information corresponding to the target object to the terminal used by the candidate objects to be invited, or the server directly sends the invitation information corresponding to the target object to the terminal used by the candidate objects to be invited.

[0132] In a game, the teaming mechanism is a common interactive mechanism in the game. Refer to Figure 4 In the human-computer interaction interface 401, an invitation friend entry 402 is displayed. In response to the triggering operation of the teaming player on the invitation friend entry 402, a teaming link is generated and sent to the friend object. In response to the confirmation operation of the friend object, the friend object joins the team of the teaming player.

[0133] In some embodiments, the player creates a team as the team leader and then sends the team link to other game players. When other game players join, the teaming is completed. The technical key point of the embodiments of this application is that after constructing the graph structure, embedding processing is performed through the line algorithm and the deepwalk algorithm. Considering the depth and width of the graph structure comprehensively, the relationship value of each node in the graph structure is calculated. The node with a large relationship value is the player with a high willingness to complete teaming. Then, these players with a high willingness to team are invited to complete the teaming, effectively solving the problem of screening the teaming population, being able to accurately screen players with the willingness to team, and reducing the disturbance to players with a low willingness to team.

[0134] The data processing method based on artificial intelligence provided by the embodiments of this application is mainly divided into three parts: 1. Constructing a network, 2. The vectorization process based on the line algorithm, and 3. The vectorization process based on the deepwalk algorithm.

[0135] First, the process of constructing a network is introduced below.

[0136] The association relationships within the virtual scene mainly include: friend relationship, conversation relationship, sworn brotherhood relationship, master-apprentice relationship, clan relationship, etc.

[0137] For the object relationship graph structure (friend network) of the friend relationship: The players with friend relationships in the game are grouped into the corresponding object relationship graph structure of the friend relationship. Select the players who have had friend relationships in the game for a period of time to construct the object relationship graph structure of the friend relationship. Among them, the nodes in the object relationship graph structure are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a friend relationship between the players corresponding to the two connected nodes.

[0138] For the object relationship graph structure (conversation network) of the conversation relationship: The players with conversation behaviors in the game are grouped into the corresponding object relationship graph structure of the conversation relationship. Select the players who have had conversation behaviors in the game for a period of time to construct the object relationship graph structure of the conversation relationship. The nodes in the object relationship graph structure are player account identifiers, and the associated edges in the object relationship graph structure indicate that there is a conversation relationship between the players corresponding to the two connected nodes. The weight of the associated edge can be positively correlated with the number of conversations.

[0139] Object relationship graph structure for sworn-brotherhood relationships (sworn-brotherhood network): The players in the game with sworn-brotherhood relationships are grouped into an object relationship graph structure corresponding to the sworn-brotherhood relationship. Select the players who had sworn-brotherhood relationships in the game over a period of time to construct an object relationship graph structure corresponding to the sworn-brotherhood relationship. The nodes in the object relationship graph structure are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a sworn-brotherhood relationship between the players corresponding to the two connected nodes.

[0140] Object relationship graph structure for master-disciple relationships (master-disciple network): The players in the game with master-disciple relationships are grouped into an object relationship graph structure corresponding to the master-disciple relationship. Select the players who had master-disciple relationships in the game over a period of time to construct an object relationship graph structure corresponding to the master-disciple relationship. The nodes in the object relationship graph structure are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a master-disciple relationship between the players corresponding to the two connected nodes.

[0141] Object relationship graph structure for team relationships (team network): The players in the game with team relationships are grouped into an object relationship graph structure corresponding to the team relationship. Select the players who had team relationships in the game over a period of time to construct an object relationship graph structure corresponding to the team relationship. The nodes in the object relationship graph structure are player account identifiers, and the associated edges in the object relationship graph structure indicate that there was a team relationship between the players corresponding to the two connected nodes.

[0142] Thus, the construction of the object relationship graph structures for each relationship dimension is completed.

[0143] Next, the vectorization process based on the LINE algorithm is introduced.

[0144] See Figure 5 , in computer science, a graph is a data structure composed of vertices and edges. Here, the vertices (represented as Figure 5 the circles in Figure 5 ) are also called nodes, and the edges (represented as

[0145] the line segments in i , j ) are also called associated edges. A graph G can be described by a vertex set V, an edge set E, and a weight set W. The numbers on the associated edges represent the weights of the associated edges.

[0145] The LINE algorithm based on first-order similarity mainly fits the joint distribution between nodes. For any edge (v i , v j ) ∈ E, its distribution is defined by formula (6):

[0146]

[0147] where W is the sum of the weights of all associated edges in the graph structure, and w ij is the weight of the associated edge (v i , v j)The corresponding weight, is the distribution on the associated edge (v i , v j ), where v i , v j are the two nodes connected by the associated edge.

[0148] u i , u j are the vectorized representations of v i , v j after the embedding process. Define the distribution on (v i , v j ) ∈ E after the embedding process as p1(v i , v j ), see Equation (7):

[0149]

[0150] where u i , u j are the vectorized representations of node v i and node v j after the embedding process.

[0151] To maintain the first-order similarity, the above two distributions should be made as close as possible. Therefore, the objective function is defined as the divergence of the above two distributions, see Equation (8):

[0152]

[0153] where O is the divergence of the above two distributions, E is the set of associated edges, (v i , v j ) ∈ E is an arbitrary associated edge, is the distribution on an arbitrary edge (v i , v j ) ∈ E, p1(v i , v j ) is the distribution on (v i , v j ) ∈ E after the embedding process.

[0154] In the vectorization process based on the line algorithm, first, according to the nodes in the graph structure and the weight values of the associated edges, the joint distribution on the associated edges is fitted Then, the line algorithm is used to determine the joint distribution after vectorization. The third step is to use the joint distribution and the joint distribution after vectorization to calculate the divergence between two nodes. The smaller the divergence value, the more similar the two nodes are in the network. The vectorization process is as follows Figure 6As shown in the figure. Embedding refers to converting nodes into multi-dimensional space vectors. For example, 100 features can be mapped to a 100-dimensional vector, with each feature corresponding to one dimension of the vector. When a feature has a value, the corresponding dimension of the vector is set to 1; otherwise, it is 0.

[0155] Due to the object relationship graph structure with multiple relationship dimensions, for each node, the vectorized representations of the node in multiple object relationship graph structures are weighted and fused to obtain the embedding feature (the first relationship feature) of the node.

[0156] The following introduces the vectorization process based on the deepwalk algorithm. The deepwalk algorithm mainly consists of a random walk part and a generating vector representation part.

[0157] See Figure 7 , use the random walk algorithm to extract a node sequence from the object relationship graph structure; the so-called random walk is to continuously randomly select a walking path on the object relationship graph structure, and finally form a path that runs through the network. Starting from a specific node, at each step of the walk, randomly select an edge connected to the current node, move along the selected edge to the next node, and repeat this process continuously. Figure 7 The trajectory 701 shown in the figure is the path obtained by the random walk.

[0158] See Figure 8 , generating vector representation is to borrow the idea of natural language processing, regard the generated node sequence as a sentence composed of words, all sequences can be regarded as a corpus, and finally use natural language processing tools to represent each node as a vector with dimension d. Since nodes themselves cannot be calculated, a mapping function φ is introduced. Here, the mapping function φ is actually a matrix, with a total of V * d parameters, where V is the number of nodes. Figure 8 The sequence used for learning in j [ (node 1) is mapped to the vector representation φ(v j [ ) of node v j [ .

[0159] See Figure 9 , through Figure 9 The conditional probability is calculated by the Huffman maximum likelihood function shown, so as to use the calculated conditional probability to optimize and learn the mapping function φ, so that the learned vectorized representation can characterize the node. Using the Huffman maximum likelihood function converts the context problem in natural language processing into the probability of walking from node 1 to leaf nodes (node 5 and node 3) on the path based on the vector representation obtained Figure 8 (each leaf node represents a node, forming a complete binary tree).

[0160] After obtaining the embedding features of each node through the DeepWalk algorithm, the embedding features are input into the sampling function to obtain the vectorized results. Here, for node A, all candidate nodes of this node A will be obtained (for example, 10). The candidate nodes can be the first-order adjacent nodes of node A. Positive and negative samples are sampled from the 10 candidate nodes, and the embedding features of node A are calculated through formula (9):

[0161]

[0162] where v i , v j is a pair of positive samples (with associated edges), v i , v n is a pair of negative samples (without associated edges), P n (v) ~ d v , and K is the number of negative sampling times. v i is the third node, v j is the fourth node, is the fourth embedding feature of the third node, is the transpose of the fourth embedding feature of the fourth node, v n is the fifth node, is the transpose of the fourth embedding feature of the fifth node. v n ~ P n (v) indicates that the nodes are all in the vector space.

[0163] Due to the existence of object relationship graph structures with multiple relationship dimensions, for each node, the vectorized representations of this node obtained from multiple object relationship graph structures based on the DeepWalk algorithm are weighted and fused to obtain the embedding features (second relationship features) of this node.

[0164] The vectorized results obtained based on the DeepWalk algorithm (or the vectorized results obtained based on the LINE algorithm) and the object features in the game are input into the neural network perceptron to obtain the relationship values of each node.

[0165] The object features here can be obtained based on the virtual resource transfer data in the game. For example, the number of days of virtual resource transfer in a year, the total number of virtual resource transfers in a year, the mode of the single-day virtual resource transfer interval in a year, the mode of the single-day virtual resource transfer interval in a year, the average number of days of virtual resource transfer per week in a year, the number of days between the last virtual resource transfer date in a year, the maximum number of consecutive virtual resource transfer days in a year, the total number of monthly virtual resource transfers in a month, etc.

[0166] The calculation logic of the neural network perceptron is as follows: sum multiple input vectors, perform a non-linear calculation on the summed vector to increase the non-linearity of the neural network perceptron. Refer to formula (10):

[0167] y = relu(relu(XW1 + bias1)W2 + bias2) (10);

[0168] Among them, X is the sum result of summing multiple input vectors, W1, bias1, W2, and bias2 are trainable parameters, and y is the teaming probability.

[0169] By using the embodiment of the present application to select players for teaming invitations, the consent rate of players agreeing to team up is relatively high, and the interaction ability of the players participating in the team is stronger.

[0170] It can be understood that in the embodiment of the present application, relevant data such as user information is involved. When the embodiment of the present application is applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards in relevant countries and regions.

[0171] In the present application, when implementing the data scraping technical solution, when the above embodiments of the present application are applied to specific products or technologies, the process of collecting, using, and processing relevant data should comply with the requirements of national laws and regulations, conform to the principles of legality, legitimacy, and necessity, do not involve obtaining data types prohibited or restricted by laws and regulations, and will not interfere with the normal operation of the target website.

[0172] Next, continue to describe the exemplary structure of the software module implementation of the artificial intelligence-based data processing device 455 provided by the embodiment of the present application. In some embodiments, such as Figure 2As shown in the figure, the software modules stored in the artificial intelligence-based data processing device 455 of the memory 450 may include: an acquisition module 4551, configured to acquire relationship data between a plurality of objects, and generate an object relationship graph structure based on the relationship data, where the plurality of objects include a target object and at least one candidate object; an embedding module 4552, configured to perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain a second relationship feature of each candidate object; an object module 4553, configured to acquire an object feature of each candidate object; a probability module 4554, configured to generate a teaming probability of each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object; and an invitation module 4555, configured to determine a candidate object to be invited for the target object based on the teaming probability of each candidate object, and perform an invitation operation for the target object on the candidate object to be invited.

[0173] In some embodiments, the acquisition module 4551 is further configured to perform at least one of the following processes: acquire social relationship data between the plurality of objects; acquire conversation relationship data between the plurality of objects; acquire role relationship data between the plurality of objects; acquire teaming relationship data between the plurality of objects; and generate the relationship data based on at least one of the social relationship data, the conversation relationship data, the role relationship data, and the teaming relationship data.

[0174] In some embodiments, the relationship data includes relationship data of at least one relationship dimension; the acquisition module 4551 is further configured to, for the relationship data of each relationship dimension, perform the following processes: generate a node corresponding to each object based on each object; when the relationship data of the relationship dimension indicates that there is a relationship corresponding to the relationship dimension between a first object and a second object, generate an association edge between the nodes respectively corresponding to the first object and the second object; and construct an object relationship graph structure corresponding to the relationship dimension based on the nodes and the association edges.

[0175] In some embodiments, the object relationship graph structure includes an object relationship graph structure corresponding to each of the relationship dimensions; the embedding module 4552 is further configured to: for the object relationship graph structure of each relationship dimension, perform a first embedding process on the object relationship graph structure of the relationship dimension based on a graph structure similarity mechanism to obtain a first relationship feature corresponding to each candidate object for the relationship dimension; when the number of relationship dimensions is multiple, for each candidate object, perform a fusion process on the first relationship features corresponding to the multiple relationship dimensions of the candidate object to obtain a first relationship feature of the candidate object; when the number of relationship dimensions is one, for each candidate object, use the first relationship feature corresponding to the relationship dimension of the candidate object as the first relationship feature of the candidate object.

[0176] In some embodiments, the object relationship graph structure of the relationship dimension includes nodes corresponding to each object and association edges between the multiple nodes, and the association edges represent that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; the embedding module 4552 is further configured to: obtain a first probability distribution of each association edge in the object relationship graph structure of the relationship dimension, and perform vector initialization processing on each node in the object relationship graph structure of the relationship dimension to obtain a first embedding feature of each node; based on the first embedding feature of each node, determine a second probability distribution of each association edge; based on the first probability distribution of each association edge and the second probability distribution of each association edge, determine a first similarity loss; update the first embedding feature of each node based on the first similarity loss to obtain a second embedding feature of each node; for each candidate object, use the second embedding feature of the node corresponding to the candidate object as the first relationship feature corresponding to the relationship dimension of the candidate object.

[0177] In some embodiments, the embedding module 4552 is further configured to: obtain the weight of each association edge; perform a summation process on the weights of all association edges of the object relationship graph structure to obtain a first summation result; use the ratio of the weight of each association edge to the first summation result as the first probability distribution of each association edge.

[0178] In some embodiments, the embedding module 4552 is further configured to perform the following processing for each association edge: perform a dot product process on the first embedding feature of the first node and the first embedding feature of the second node to obtain a first dot product result, where the first node and the second node are the two nodes connected by the association edge; perform a probability function mapping process on the first dot product result to obtain the second probability distribution of the association edge.

[0179] In some embodiments, the embedding module 4552 is further configured to perform the following processing for each of the associated edges: perform a logarithmic processing on the second probability distribution to obtain a first logarithmic result; multiply the first logarithmic result by the first probability distribution to obtain a first multiplication result; perform a summation processing on the first multiplication results corresponding to multiple associated edges to obtain a second summation result, and use the opposite number of the first summation result as the first similarity loss.

[0180] In some embodiments, the object relationship graph structure includes an object relationship graph structure corresponding to each of the relationship dimensions; the embedding module 4552 is further configured to: for the object relationship graph structure of each relationship dimension, perform a second embedding processing based on a random walk mechanism on the object relationship graph structure of the relationship dimension to obtain a second relationship feature of each candidate object corresponding to the relationship dimension; when the number of relationship dimensions is multiple, for each candidate object, perform a fusion processing on the second relationship features of the candidate object corresponding to multiple relationship dimensions to obtain a second relationship feature of the candidate object; when the number of relationship dimensions is one, for each candidate object, use the second relationship feature of the candidate object corresponding to the relationship dimension as the second relationship feature of the candidate object.

[0181] In some embodiments, the object relationship graph structure of the relationship dimension includes nodes corresponding to each object and associated edges between multiple nodes, and the associated edges represent that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; the embedding module 4552 is further configured to: perform a random walk processing based on the associated edges on the object relationship graph structure of the relationship dimension to obtain multiple node sequences; sequentially perform the following processing for each node sequence: perform an initialization processing on each node in the node sequence to obtain a third embedding feature of each node in the node sequence, and based on the node sequence, perform an update processing on the third embedding feature of each node in the node sequence to obtain a fourth embedding feature of each node in the node sequence; perform a sampling processing based on the fourth embedding feature of each node to obtain a sampled embedding feature of each node; for each candidate object, use the sampled embedding feature of the node corresponding to the candidate object as the second relationship feature of the candidate object corresponding to the relationship dimension.

[0182] In some embodiments, the embedding module 4552 is further configured to perform the following processing for each of the nodes: when the node has not been subjected to the initialization processing, perform a random initialization processing on the node to obtain a third embedding feature of the node; when the node has been subjected to the initialization processing, use the fourth embedding feature obtained by the most recent update of the node as the third embedding feature of the node.

[0183] In some embodiments, the embedding module 4552 is further configured to perform the following processing for each of the nodes: perform a context node prediction processing on the node sequence based on the third embedding feature of the node to obtain a node prediction result; update the third embedding feature of the node based on the error between the real node represented by the node sequence and the node prediction result to obtain a fourth embedding feature of the node.

[0184] In some embodiments, the embedding module 4552 is further configured to perform the following processing for each of the nodes in the object relationship graph structure: obtain candidate nodes adjacent to the node; obtain a third node and a fourth node having a connection relationship from the candidate nodes, and obtain a third node and a fifth node not having the connection relationship from the candidate nodes; determine a first transposed result of the fourth embedding feature of the fourth node and a second transposed result of the fourth embedding feature of the fifth node; determine a first inner product between the fourth embedding feature of the third node and the first transposed result, and a second inner product between the fourth embedding feature of the third node and the second transposed result; obtain an embedding feature that is positively correlated with the first inner product and negatively correlated with the second inner product as the sampling embedding feature of the node.

[0185] In some embodiments, the probability module 4554 is further configured to perform the following processing for each of the candidate objects: perform a mapping processing on the first relationship feature of the candidate object and the second relationship feature of the candidate object to obtain a comprehensive relationship feature of the candidate object; perform a pooling fusion processing on the comprehensive relationship feature of the candidate object and the object feature of the candidate object to obtain a pooling fusion feature of the candidate object; perform a multi-layer perception processing on the candidate object to obtain a teaming probability of the candidate object.

[0186] An embodiment of the present application provides a computer program product, which includes computer executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer executable instructions from the computer-readable storage medium, and the processor executes the computer executable instructions, so that the electronic device executes the above-mentioned artificial intelligence-based data processing method of the embodiment of the present application.

[0187] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, where the computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor will be caused to execute the data processing method based on artificial intelligence provided by the embodiment of the present application. For example, as Figures 3A - 3C shown in the data processing method based on artificial intelligence.

[0188] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0189] In some embodiments, the computer-executable instructions may be in the form of a program, software, software module, script, or code, and may be written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including being deployed as an independent program or being deployed as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0190] As an example, the computer-executable instructions may or may not correspond to a file in the file system, and may be stored as part of a file that stores other programs or data. For example, they may be stored in one or more scripts in a HyperText Markup Language (HTML) document, stored in a single file dedicated to the program under discussion, or stored in multiple cooperating files (for example, files that store one or more modules, subroutines, or code portions).

[0191] As an example, the computer-executable instructions may be deployed to execute on one electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed at multiple locations and interconnected through a communication network.

[0192] In summary, through the embodiments of the present application, relationship data between multiple objects is obtained, and an object relationship graph structure is generated based on the relationship data. The multiple objects include a target object and at least one candidate object. Here, the relationship data is converted into a graph structure form. The object relationship graph structure is subjected to a first embedding process based on a graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and the object relationship graph structure is subjected to a second embedding process based on a random walk mechanism to obtain the second relationship feature of each candidate object; the object feature of each candidate object is obtained from the relationship data. Here, the features of the candidate object at the relationship level are determined based on two dimensions of the graph structure similarity mechanism and the random walk mechanism respectively. Based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object, the teaming probability of each candidate object is generated. The teaming probability is determined by combining the features of the relationship dimension and the object dimension. Based on the teaming probability of each candidate object, the candidate object to be invited for the target object is determined, and an invitation operation corresponding to the target object is performed on the candidate object to be invited. Since the teaming probability takes into account the features of both the relationship dimension and the object dimension, the candidate object to be invited determined based on the teaming probability has a relatively high possibility of accepting the invitation to form a team, which can improve the success rate of the invitation operation and thus improve the human-computer interaction efficiency.

[0193] As described above, the above are only the embodiments of the present application and are not intended to limit the protection scope of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A data processing method based on artificial intelligence, characterized in that, The method includes: Obtaining relationship data among multiple objects, and generating an object relationship graph structure based on the relationship data, where the multiple objects include a target object and at least one candidate object; Performing a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object, and performing a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain a second relationship feature of each candidate object; Obtaining an object feature of each candidate object; Generating a teaming probability of each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object; Determining candidate objects to be invited for the target object based on the teaming probability of each candidate object, and performing an invitation operation for the target object on the candidate objects to be invited.

2. The method according to claim 1, wherein The relationship data includes relationship data of at least one relationship dimension, and the object relationship graph structure includes an object relationship graph structure of at least one of the relationship dimensions; The generating an object relationship graph structure based on the relationship data includes: Performing the following processing for the relationship data of each relationship dimension; Generating a node corresponding to each object based on each object; When the relationship data of the relationship dimension indicates that there is a relationship corresponding to the relationship dimension between a first object and a second object, generating an association edge between the nodes respectively corresponding to the first object and the second object, where the first object and the second object are from the multiple objects; Constructing an object relationship graph structure corresponding to the relationship dimension based on the node and the association edge.

3. The method according to claim 1, characterized in that, The object relationship graph structure includes an object relationship graph structure corresponding to each relationship dimension; The performing a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object includes: For the object relationship graph structure of each relationship dimension, performing a first embedding process on the object relationship graph structure of the relationship dimension based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object corresponding to the relationship dimension; When the number of relationship dimensions is multiple, for each candidate object, performing a fusion process on the first relationship features of the candidate object corresponding to the multiple relationship dimensions to obtain the first relationship feature of the candidate object; When the number of relationship dimensions is one, for each candidate object, using the first relationship feature of the candidate object corresponding to the relationship dimension as the first relationship feature of the candidate object.

4. The method according to claim 3, wherein The object relationship graph structure of the relationship dimension includes nodes corresponding to each object and association edges between the multiple nodes, and the association edge indicates that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; The performing a first embedding process on the object relationship graph structure of the relationship dimension based on a graph structure similarity mechanism to obtain a first relationship feature of each candidate object corresponding to the relationship dimension includes: Obtain the first probability distribution of each of the associated edges in the object relationship graph structure of the relationship dimension, and perform vector initialization processing on each of the nodes in the object relationship graph structure of the relationship dimension to obtain the first embedding feature of each of the nodes; Based on the first embedding feature of each of the nodes, determine the second probability distribution of each of the associated edges; Based on the first probability distribution of each of the associated edges and the second probability distribution of each of the associated edges, determine the first similarity loss; Update the first embedding feature of each of the nodes based on the first similarity loss to obtain the second embedding feature of each of the nodes; For each of the candidate objects, use the second embedding feature of the node corresponding to the candidate object as the first relationship feature of the candidate object corresponding to the relationship dimension.

5. The method according to claim 4, wherein The obtaining the first probability distribution of each of the associated edges in the object relationship graph structure of the relationship dimension includes: Obtain the weight of each of the associated edges in the object relationship graph structure of the relationship dimension; Perform a summation process on the weights of all the associated edges in the object relationship graph structure of the relationship dimension to obtain a first summation result; Use the ratio of the weight of each of the associated edges to the first summation result as the first probability distribution of each of the associated edges.

6. The method according to claim 4, wherein The determining the second probability distribution of each of the associated edges based on the first embedding feature of each of the nodes includes: Perform the following process for each of the associated edges: Perform a dot product process on the first embedding feature of the first node and the first embedding feature of the second node to obtain a first dot product result, where the first node and the second node are the two nodes connected by the associated edge; Perform a probability function mapping process on the first dot product result to obtain the second probability distribution of the associated edge.

7. The method according to claim 4, characterized in that, The determining the first similarity loss based on the first probability distribution of each of the associated edges and the second probability distribution of each of the associated edges includes: Perform the following process for each of the associated edges: Perform a logarithm process on the second probability distribution to obtain a first logarithm result; Multiply the first logarithm result by the first probability distribution to obtain a first multiplication result; Perform a summation process on the first multiplication results corresponding to multiple associated edges to obtain a second summation result, and use the negative value of the first summation result as the first similarity loss.

8. The method according to claim 1, wherein The object relationship graph structure includes an object relationship graph structure corresponding to each of the relationship dimensions; The performing a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain the second relationship feature of each of the candidate objects includes: For the object relationship graph structure of each of the relationship dimensions, perform a second embedding process on the object relationship graph structure of the relationship dimension based on a random walk mechanism to obtain the second relationship feature of each of the candidate objects corresponding to the relationship dimension; When the number of relationship dimensions is multiple, for each of the candidate objects, perform a fusion process on the second relationship features of the candidate object corresponding to multiple relationship dimensions to obtain the second relationship feature of the candidate object; When the number of the relationship dimensions is one, for each of the candidate objects, the second relationship feature corresponding to the relationship dimension of the candidate object is used as the second relationship feature of the candidate object.

9. The method according to claim 8, wherein The object relationship graph structure of the relationship dimension includes nodes corresponding to each of the objects and association edges between the multiple nodes, and the association edges represent that there is a relationship corresponding to the relationship dimension between the objects corresponding to the two connected nodes; The second embedding process based on the random walk mechanism for the object relationship graph structure of the relationship dimension to obtain the second relationship feature corresponding to the relationship dimension of each candidate object includes: Performing a random walk process based on the association edges on the object relationship graph structure of the relationship dimension to obtain a plurality of node sequences; Successively performing the following process for each of the node sequences: performing an initialization process on each of the nodes in the node sequence to obtain a third embedding feature of each of the nodes in the node sequence, and updating the third embedding feature of each of the nodes in the node sequence based on the node sequence to obtain a fourth embedding feature of each of the nodes in the node sequence; Performing a sampling process based on the fourth embedding feature of each of the nodes to obtain a sampled embedding feature of each of the nodes; For each of the candidate objects, the sampled embedding feature of the node corresponding to the candidate object is used as the second relationship feature corresponding to the relationship dimension of the candidate object.

10. The method according to claim 9, characterized in that, The performing an initialization process on each of the nodes in the node sequence to obtain a third embedding feature of each of the nodes in the node sequence includes: Performing the following process for each of the nodes: When the node has not been subjected to the initialization process, performing a random initialization process on the node to obtain a third embedding feature of the node; When the node has been subjected to the initialization process, using the fourth embedding feature obtained by the most recent update of the node as the third embedding feature of the node.

11. The method according to claim 9, wherein The updating the third embedding feature of each of the nodes in the node sequence based on the node sequence to obtain a fourth embedding feature of each of the nodes in the node sequence includes: Performing the following process for each of the nodes: Performing a context node prediction process on the node sequence based on the third embedding feature of the node to obtain a node prediction result; Updating the third embedding feature of the node based on the error between the real nodes represented by the node sequence and the node prediction result to obtain a fourth embedding feature of the node.

12. The method according to claim 9, wherein The performing a sampling process based on the fourth embedding feature of each of the nodes to obtain a sampled embedding feature of each of the nodes includes: Performing the following process for each of the nodes in the object relationship graph structure: Obtaining candidate nodes adjacent to the node; Obtaining a third node and a fourth node having a connection relationship from the candidate nodes, and obtaining a third node and a fifth node not having the connection relationship from the candidate nodes; Determine the first transposed result of the fourth embedding feature of the fourth node and the second transposed result of the fourth embedding feature of the fifth node; Determine the first inner product of the fourth embedding feature of the third node and the first transposed result, and the second inner product of the fourth embedding feature of the third node and the second transposed result; Obtain the embedding feature that is positively correlated with the first inner product and negatively correlated with the second inner product as the sampled embedding feature of the node.

13. The method according to claim 1, characterized in that, The generating the teaming probability for each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object includes: Perform the following processing for each candidate object: Perform a mapping process on the first relationship feature of the candidate object and the second relationship feature of the candidate object to obtain the comprehensive relationship feature of the candidate object; Perform a pooling fusion process on the comprehensive relationship feature of the candidate object and the object feature of the candidate object to obtain the pooling fusion feature of the candidate object; Perform a multi-layer perception process on the candidate object to obtain the teaming probability of the candidate object.

14. An artificial intelligence-based data processing device, characterized in that, The apparatus includes: An acquisition module, configured to acquire relationship data between multiple objects and generate an object relationship graph structure based on the relationship data, where the multiple objects include a target object and at least one candidate object; An embedding module, configured to perform a first embedding process on the object relationship graph structure based on a graph structure similarity mechanism to obtain the first relationship feature of each candidate object, and perform a second embedding process on the object relationship graph structure based on a random walk mechanism to obtain the second relationship feature of each candidate object; An object module, configured to acquire the object feature of each candidate object; A probability module, configured to generate the teaming probability of each candidate object based on the first relationship feature of each candidate object, the second relationship feature of each candidate object, and the object feature of each candidate object; An invitation module, configured to determine the candidate objects to be invited for the target object based on the teaming probability of each candidate object, and perform an invitation operation for the target object on the candidate objects to be invited.

15. An electronic device, characterized in that, The electronic device includes: A memory, configured to store computer-executable instructions; A processor, configured to implement the artificial intelligence-based data processing method according to any one of claims 1 to 13 when executing the computer-executable instructions stored in the memory.

16. A computer-readable storage medium stores computer-executable instructions, characterized in that, The computer-executable instructions, when executed by the processor, implement the artificial intelligence-based data processing method according to any one of claims 1 to 13.

17. A computer program product, comprising computer-executable instructions, characterized in that, The computer-executable instructions, when executed by the processor, implement the artificial intelligence-based data processing method according to any one of claims 1 to 13.

Images