CN111563802B

CN111563802B - Virtual article recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN111563802B
Application number: CN202010388740.2A
Authority: CN
Inventors: 钱宇秋; 陈思
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-05-09
Filing date: 2020-05-09
Publication date: 2023-04-14
Anticipated expiration: 2040-05-09
Also published as: CN111563802A

Abstract

The invention provides a virtual article recommendation method which comprises the following steps: determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user; determining a sequence subgraph corresponding to a real purchase sequence and a pseudo prop sequence of a target user; determining corresponding sequence segment feature vectors; sharing the sequence segment characteristic vectors to form new sequence segment characteristic vectors; and determining the virtual articles to be purchased by the target user so as to recommend the virtual articles to the target user. The invention also provides a virtual article recommendation device, electronic equipment and a storage medium. The virtual article recommendation method and the virtual article recommendation system can accurately recommend the virtual article to the user by using the virtual article recommendation model, reduce the time for the user to search the virtual article, and improve the use experience of the user.

Description

Virtual article recommendation method and device, electronic equipment and storage medium

Technical Field

The present invention relates to information recommendation technologies, and in particular, to a virtual item recommendation method and apparatus, an electronic device, and a storage medium.

Background

In the prior art, a game user usually purchases an interesting virtual object (e.g. a game item) in a virtual environment, wherein the item prop (PROPS) generally refers to any movable object for decoration and arrangement in a scene. There are also items in the game that are convenient items for the game target user, which are usually acquired by the game target user for a task or by purchase. In the process, different game target users often have different prop requirements at different times, and the required props are accurately recommended to the game target users, so that the purchase rate of the game props by the game target users can be improved, the game revenue of game application merchants can be increased, and the decrease of the target user pair stickiness caused by inaccurate prop recommendation of the game target users can be reduced.

Disclosure of Invention

In view of this, an embodiment of the present invention provides a virtual item recommendation method, an apparatus, an electronic device, and a storage medium, where a technical solution of an embodiment of the present invention is implemented as follows:

the invention provides a virtual article recommendation method, which comprises the following steps:

responding to the virtual article purchase request, and obtaining the virtual article purchase behavior parameter information of a target user;

determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user;

determining corresponding sequence segment feature vectors based on the pseudo item tasks and the real purchase sequences of the target users;

sharing the sequence segment characteristic vectors to form new sequence segment characteristic vectors;

and determining the virtual articles to be purchased by the target user based on the user feature vector of the target user so as to realize recommendation of the virtual articles to the target user.

In the foregoing solution, the acquiring of the virtual article purchasing behavior parameter information of the target user includes:

acquiring original data sets of different dimensions corresponding to the target user;

determining a purchase sequence and corresponding purchase time of the virtual article according to the original data sets with different dimensions;

and determining the target user virtual article purchasing behavior parameters in different unit time based on the purchasing sequence of the pseudo-articles and the corresponding purchasing time.

An embodiment of the present invention further provides a virtual item recommendation apparatus, including:

the information transmission module is used for responding to the virtual article purchasing request, and acquiring the virtual article purchasing behavior parameter information of the target user;

the information processing module is used for determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user;

the information processing module is used for determining corresponding sequence segment characteristic vectors based on the pseudo item tasks and the real purchase sequences of the target users;

the information processing module is used for sharing the sequence segment feature vectors to form new sequence segment feature vectors;

the information processing module is used for determining the virtual articles to be purchased by the target user through the virtual article recommendation model based on the user feature vector of the target user so as to realize the recommendation of the virtual articles to the target user.

In the above-mentioned scheme, the first and second light sources,

the information processing module is used for determining a real purchasing sequence matched with the target user based on the virtual article purchasing behavior parameter information of the target user;

the information processing module is used for determining a real purchase sequence of the target user and a sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchase sequence of the target user;

and the information processing module is used for determining the sequence segment feature vector corresponding to the target user through the sequence subgraph.

In the above-mentioned scheme, the first step of the method,

the information processing module is used for determining a real purchase sequence and a pseudo item sequence based on the virtual article purchase behavior parameter information of the target user;

the information processing module is used for determining nodes of the sequence subgraph by determining nodes of a target virtual article in the real purchase sequence and nodes in the pseudo item sequence;

the information processing module is used for determining a first type edge of the sequence subgraph based on the time sequence of the target virtual article in the real purchasing sequence;

the information processing module is used for determining a second type edge of the sequence subgraph based on the time sequence of the target virtual article in the pseudo prop sequence;

the information processing module is used for determining a third type edge of the sequence subgraph based on the incidence relation between the real purchasing sequence and the pseudo prop sequence;

and the information processing module is used for determining the sequence subgraph matched with the virtual article purchasing behavior of the target user according to the nodes of the sequence subgraph, the first type edge, the second type edge and the third type edge.

In the above-mentioned scheme, the first step of the method,

the information processing module is used for linearly weighting all feature nodes of the sequence subgraph and determining a first sequence segment feature vector, wherein the first sequence segment feature vector is used for representing the purchasing behavior preference of the target user;

the information processing module is used for extracting the purchase parameters of the last quasi-item of the target user and determining a second sequence segment feature vector;

and the information processing module is used for splicing the first sequence segment feature vector and the second sequence segment feature vector to determine a corresponding sequence segment feature vector.

In the above-mentioned scheme, the first step of the method,

the information processing module is used for initializing target user vectors shared by different sequences in the sequence subgraphs matched with the virtual article purchasing behaviors of the target users;

the information processing module is used for processing the target user vector through a first user neural network model in the virtual article recommendation model and determining a first user single sequence feature matched with a real purchase sequence;

the information processing module is used for processing the target user vector through a second user neural network model in the virtual article recommendation model and determining a second user single-sequence feature matched with the pseudo item sequence;

and the information processing module is used for splicing the first user single-sequence feature and the second user single-sequence feature to form a user feature vector of the target user.

In the above-mentioned scheme, the first step of the method,

the information processing module is used for determining a multitask loss function matched with the virtual article recommendation model;

the information processing module is used for adjusting network parameters of the virtual article recommendation model based on the sequence segment feature vector, the user feature vector of the target user and the multitask loss function;

the information processing module is used for processing the virtual article recommendation model until loss functions of different dimensions corresponding to the virtual article recommendation model reach corresponding convergence conditions; so as to realize the matching of the virtual item recommendation model and the target user.

In the above-mentioned scheme, the first step of the method,

the information processing module is used for sequencing the recall sequence of the virtual articles based on the names of the virtual articles purchased by the target user and the categories of the virtual articles to be purchased;

and the information processing module is used for adjusting the sequence of the virtual articles displayed in the user interface according to the sorting result of the recall sequence of the virtual articles.

In the above-mentioned scheme, the first and second light sources,

the information processing module is used for determining game roles held by the target user when the virtual object is a game prop, wherein different game roles are matched with the corresponding game props;

the information processing module is used for determining a purchase sequence and a corresponding purchase time of the corresponding game prop based on the game role;

and the information processing module is used for determining the purchasing behavior parameters of the target user virtual articles in different unit time based on the purchasing sequence and the corresponding purchasing time of the game props.

An embodiment of the present invention further provides an electronic device, where the electronic device includes:

a memory for storing executable instructions;

and the processor is used for realizing the virtual article recommendation model training method in the preamble or realizing the preamble virtual article recommendation method when the executable instructions stored in the memory are run.

An embodiment of the present invention further provides a computer-readable storage medium, which stores executable instructions, and when the executable instructions are executed by a processor, the method for training a virtual article recommendation model in the preamble is implemented, or the method for recommending the virtual article in the preamble is implemented.

The embodiment of the invention has the following beneficial effects:

the embodiment of the invention acquires the virtual article purchasing behavior parameter information of a target user; determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user; determining a real purchase sequence of the target user and a sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchase sequence of the target user; determining corresponding sequence segment feature vectors based on a graph neural network model in the virtual article recommendation model through the sequence subgraphs; sharing the sequence segment feature vectors through a self-attention learning mechanism of the virtual article recommendation model to form new sequence segment feature vectors; based on the user feature vector of the target user, the name of the virtual article to be purchased by the target user and the category of the virtual article to be purchased can be determined through the virtual article recommendation model, and further the virtual article can be accurately recommended to the user by using the virtual article recommendation model, so that the time for the user to search for the virtual article is reduced, and the use experience of the user is improved.

Drawings

Fig. 1 is a schematic view of a use scenario of a virtual item recommendation model training method according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a virtual item recommendation device according to an embodiment of the present invention;

fig. 3 is an optional flowchart of the training method for the virtual item recommendation model according to the embodiment of the present invention;

fig. 4 is an alternative flow chart of a virtual item recommendation method according to an embodiment of the present invention;

fig. 5 is an alternative flowchart of a virtual item recommendation method according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a front end display of a virtual article according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a front end display of a virtual article according to an embodiment of the present invention;

fig. 8 is an alternative flowchart of a training method for a virtual item recommendation model according to an embodiment of the present invention;

fig. 9 is a schematic diagram of data flow in a training process of a virtual item recommendation model according to an embodiment of the present invention;

fig. 10 is a schematic diagram of a sequence segment subgraph in a training process of a virtual item recommendation model according to an embodiment of the present invention;

fig. 11 is a schematic diagram of a sequence segment self-attention mechanics learning process in a training process of a virtual item recommendation model according to an embodiment of the present invention;

fig. 12 is a schematic diagram illustrating prediction performed by sequence segments and user characteristics in a training process of a virtual item recommendation model according to an embodiment of the present invention;

fig. 13 is a schematic view of processing sequence segments in the process of recommending game items by using the virtual item recommendation model according to the embodiment of the present invention;

fig. 14 is a schematic diagram of a front end display of a virtual article subjected to display sequence adjustment according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings, the described embodiments should not be construed as limiting the present invention, and all other embodiments that can be obtained by a person skilled in the art without making creative efforts fall within the protection scope of the present invention.

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

Before further detailed description of the embodiments of the present invention, terms and expressions mentioned in the embodiments of the present invention are explained, and the terms and expressions mentioned in the embodiments of the present invention are applied to the following explanations.

1) In response to the condition or state indicating that the executed operation depends on, one or more of the executed operations may be in real-time or may have a set delay when the dependent condition or state is satisfied; there is no restriction on the order of execution of the operations performed unless otherwise specified.

2) Based on the condition or state on which the operation to be performed depends, the operation or operations to be performed may be in real time or may have a set delay when the condition or state on which the operation depends is satisfied; there is no restriction on the order of execution of the operations performed unless otherwise specified.

3) Convolutional Neural Networks (CNN Convolutional Neural Networks) are a class of Feed-forward Neural Networks (Feed-forward Neural Networks) that contain convolution computations and have a deep structure, and are one of the representative algorithms of deep learning (deep learning). The convolutional neural network has a representation learning (representation learning) capability, and can perform shift-invariant classification (shift-invariant classification) on input information according to a hierarchical structure of the convolutional neural network.

4) And (4) model training, namely performing multi-classification learning on the image data set. The model can be constructed by adopting deep learning frames such as Tensor Flow, torch and the like, and a multi-classification model is formed by combining multiple layers of neural network layers such as CNN and the like. The input of the model is a three-channel or original channel matrix formed by reading an image through openCV and other tools, the output of the model is multi-classification probability, and the webpage category is finally output through softmax and other algorithms. During training, the model approaches to a correct trend through target functions such as cross entropy and the like.

5) Neural Networks (NN): an Artificial Neural Network (ANN), referred to as Neural Network or Neural Network for short, is a mathematical model or computational model that imitates the structure and function of biological Neural Network (central nervous system of animals, especially brain) in the field of machine learning and cognitive science, and is used for estimating or approximating functions.

6) Graph Neural Network (GNN): a neural network acting directly on a graph structure mainly processes data of a non-Euclidean space structure (graph structure). Have an input order that ignores nodes; in the calculation process, the representation of the node is influenced by the neighbor nodes around the node, and the connection of the graph is unchanged; the representation of graph structure enables graph-based reasoning. In general, a graph neural network consists of two modules: the system comprises a Propagation Module (Propagation Module) and an Output Module (Output Module), wherein the Propagation Module is used for transmitting information between nodes in the graph and updating the state, and the Output Module is used for defining an objective function according to different tasks based on vector representation of the nodes and edges of the graph. The graph neural network has: graph Convolutional Neural Networks (GCNs), gated Graph Neural Networks (GGNNs), and Graph Attention Neural Networks (GAT) based on Attention-based mechanisms.

7) Directed graph: representing the relationship from object to object, a directed graph can be represented by ordered triples (V (D), A (D), ψ D), where ψ D is a correlation function, which is an ordered pair of elements where each element in A (D) corresponds to V (D).

8) Encoder-decoder architecture: a network architecture commonly used for machine translation technology. The decoder receives the output result of the encoder as input and outputs a corresponding text sequence of another language.

9) Softmax: the normalized exponential function is a generalization of the logistic function. It can "compress" a K-dimensional vector containing arbitrary real numbers into another K-dimensional real vector such that each element ranges between [0,1] and the sum of all elements is 1.

10 Pseudo prop task: instead of a prediction task of purchasing a prop, a forgery prediction task with known results is performed in combination with game planning professional knowledge, such as predicting the type of the prop (of course, basic resources, different system props and the payment mode of the prop)

11 Multitask recommendation: and combining a plurality of prediction tasks (including a prop purchasing task and a prop type prediction task), sharing user and prop information, and making different recommendations.

12 Sequence recommendation: and analyzing the data time sequence, and recommending game items which are wanted to be purchased by a target user (a game player) at the next moment.

13 Play items): virtual items sold in the game, gift bags (including character skills and weapons), character decorations (skin), game characters (hero), and the like.

Fig. 1 is a schematic view of a usage scenario of a virtual item recommendation method provided in an embodiment of the present invention, referring to fig. 1, a terminal (including a terminal 10-1 and a terminal 10-2) is provided with a client capable of displaying software of different virtual items to be sold, for example, clients or plug-ins of different types of games or game applets, and a user can obtain and display all virtual items to be sold through the corresponding client; the terminal is connected to the server 200 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two, and uses a wireless link to realize data transmission.

As an example, the server 200 is configured to deploy the virtual article recommendation apparatus to implement the virtual article recommendation method provided by the present invention, and the latter may deploy a trained virtual article recommendation model to implement recommendation of virtual articles in different environments (for example, in a game environment or in a virtual transaction environment), specifically, before using the virtual article recommendation model, the virtual article recommendation model needs to be trained, and the specific process includes: acquiring virtual article purchasing behavior parameter information of a target user; determining a sequence subgraph matched with the virtual article purchasing behavior of the target user based on the parameter information of the virtual article purchasing behavior of the target user; determining a graph neural network model in the virtual article recommendation model through the sequence subgraphs; determining corresponding sequence segment feature vectors based on a graph neural network model in the virtual article recommendation model; and adjusting the network parameters of the virtual article recommendation model based on the sequence segment feature vectors and the user feature vectors of the target user to realize the matching of the virtual article recommendation model and the target user. Certainly, the virtual item recommendation device provided by the present invention may train based on purchasing behaviors of a same target User on virtual items in different game environments, generally select and purchase virtual items from different data sources in different game roles held by the same User, and finally present, on a User Interface (UI User Interface), a name of a virtual item purchased by the target User and a category of the virtual item to be purchased, which are determined by a virtual item recommendation model, on a User Interface (UI User Interface), and the obtained name of the virtual item purchased by the target User and the category of the virtual item to be purchased, which are determined by the virtual item recommendation model, may also be called by other application programs, and of course, the virtual item recommendation model matched with the corresponding target User may also be migrated to different virtual recommendation processes (for example, a recommendation process of a web game, a game recommendation process of a small program, or a client recommendation process of a short video game).

Certainly, after training of the virtual item recommendation model is completed, recommendation can be performed through the virtual item recommendation model, and the method specifically includes: responding to a virtual article purchasing request, and acquiring virtual article purchasing behavior parameter information of a target user; determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user; determining a real purchase sequence of the target user and a sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchase sequence of the target user; determining corresponding sequence segment feature vectors based on a graph neural network model in the virtual article recommendation model through the sequence subgraphs; sharing the sequence segment feature vectors through a self-attention learning mechanism of the virtual article recommendation model to form new sequence segment feature vectors; and determining the name of the virtual article to be purchased by the target user and the category of the virtual article to be purchased through the virtual article recommendation model based on the user feature vector of the target user so as to realize recommendation of the virtual article to the target user.

As will be described in detail below, the virtual item recommendation device according to the embodiment of the present invention may be implemented in various forms, such as a dedicated terminal with a processing function of the virtual item recommendation device, or a server with a processing function of the virtual item recommendation device, for example, the server 200 in fig. 1. Fig. 2 is a schematic diagram of a component structure of a virtual item recommendation device according to an embodiment of the present invention, and it is understood that fig. 2 only shows an exemplary structure of the virtual item recommendation device, and not a whole structure thereof, and a part of or the whole structure shown in fig. 2 may be implemented as needed.

The virtual article recommendation device provided by the embodiment of the invention comprises: at least one processor 201, memory 202, user interface 203, and at least one network interface 204. The various components in the virtual item recommendation device are coupled together by a bus system 205. It will be appreciated that the bus system 205 is used to enable communications among the components. The bus system 205 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 205 in FIG. 2.

The user interface 203 may include, among other things, a display, a keyboard, a mouse, a trackball, a click wheel, a key, a button, a touch pad, or a touch screen.

It will be appreciated that the memory 202 can be either volatile memory or nonvolatile memory, and can include both volatile and nonvolatile memory. The memory 202 in embodiments of the present invention is capable of storing data to support operation of the terminal (e.g., 10-1). Examples of such data include: any computer program, such as an operating system and application programs, for operation on a terminal, such as 10-1. The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, and the like, and is used for implementing various basic services and processing hardware-based tasks. The application program may include various application programs.

In some embodiments, the virtual article recommendation apparatus provided in the embodiments of the present invention may be implemented by a combination of hardware and software, and as an example, the virtual article recommendation apparatus provided in the embodiments of the present invention may be a processor in the form of a hardware decoding processor, which is programmed to execute the virtual article recommendation method provided in the embodiments of the present invention. For example, a processor in the form of a hardware decoding processor may employ one or more Application Specific Integrated Circuits (ASICs), DSPs, programmable Logic Devices (PLDs), complex Programmable Logic Devices (CPLDs), field Programmable Gate Arrays (FPGAs), or other electronic components.

As an example of the virtual article recommendation apparatus provided by the embodiment of the present invention implemented by combining software and hardware, the virtual article recommendation apparatus provided by the embodiment of the present invention may be directly embodied as a combination of software modules executed by the processor 201, where the software modules may be located in a storage medium, the storage medium is located in the memory 202, and the processor 201 reads executable instructions included in the software modules in the memory 202, and completes the virtual article recommendation method provided by the embodiment of the present invention in combination with necessary hardware (for example, including the processor 201 and other components connected to the bus 205).

By way of example, the Processor 201 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor or the like.

As an example of the virtual item recommendation Device provided in the embodiment of the present invention implemented by hardware, the Device provided in the embodiment of the present invention may be implemented by directly using the processor 201 in the form of a hardware decoding processor, for example, by being executed by one or more Application Specific Integrated Circuits (ASICs), DSPs, programmable Logic Devices (PLDs), complex Programmable Logic Devices (CPLDs), field Programmable Gate Arrays (FPGAs), or other electronic components, to implement the virtual item recommendation method provided in the embodiment of the present invention.

The memory 202 in embodiments of the present invention is used to store various types of data to support the operation of the virtual item recommendation device. Examples of such data include: any executable instructions for operating on the virtual item recommendation device, such as executable instructions, may be embodied in the executable instructions to implement the virtual item recommendation method of embodiments of the present invention.

In other embodiments, the virtual item recommendation apparatus provided in the embodiments of the present invention may be implemented in software, and fig. 2 illustrates the virtual item recommendation apparatus stored in the memory 202, which may be software in the form of programs, plug-ins, and the like, and includes a series of modules, as an example of the programs stored in the memory 202, the virtual item recommendation apparatus may include the following software modules:

an information transmission module 2081 and an information processing module 2082. When the software modules in the virtual item recommendation device are read into the RAM by the processor 201 and executed, the virtual item recommendation method provided by the embodiment of the invention is implemented, wherein the functions of each software module in the virtual item recommendation device include:

the information transmission module 2081 is used for the information transmission module and is used for obtaining the virtual article purchasing behavior parameter information of the target user;

the information processing module 2082 is configured to determine a pseudo item task matched with the virtual item purchasing behavior of the target user based on the parameter information of the virtual item purchasing behavior of the target user;

the information processing module 2082 is configured to determine corresponding sequence segment feature vectors based on the pseudo item task and the actual purchase sequence of the target user;

the information processing module 2082 is configured to share the sequence segment feature vectors to form new sequence segment feature vectors;

the information processing module 2082 is configured to determine, through the virtual article recommendation model, a virtual article to be purchased by the target user based on the user feature vector of the target user, so as to implement recommendation of the virtual article to the target user.

Continuing to describe the virtual item pushing method provided by the embodiment of the present invention with reference to the virtual item recommendation device shown in fig. 2, first, a training process of the virtual item recommendation model deployed in the server is described, where, referring to fig. 3, fig. 3 is an optional flowchart of the training method of the virtual item recommendation model provided by the embodiment of the present invention, it can be understood that the steps shown in fig. 3 may be executed by various electronic devices operating the virtual item recommendation device, for example, a dedicated terminal with the virtual item recommendation device, a game server, or a server cluster of a virtual item provider, where the dedicated terminal with the virtual item recommendation device may be the electronic device with the virtual item recommendation device in the embodiment shown in the foregoing fig. 2. In order to overcome the defect of inaccurate virtual article recommendation caused by the traditional recommendation mode, the technical scheme provided by the invention uses an Artificial Intelligence technology, and an Artificial Intelligence AI (Artificial Intelligence) is a theory, a method, a technology and an application system which simulate, extend and expand human Intelligence by using a digital computer or a machine controlled by the digital computer, sense the environment, acquire knowledge and obtain the best result by using the knowledge. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the implementation method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject, and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence base technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Machine Learning (ML) is a multi-domain cross subject, and relates to multi-domain subjects such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The method specially studies how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning. With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and applied in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical care, smart customer service, and the like.

The following is a detailed description of the steps shown in fig. 3.

Step 301: the virtual article recommending device acquires the virtual article purchasing behavior parameter information of the target user.

Wherein the purchasing behavior parameter information comprises: purchase sequence, purchase time, prop type, payment mode, number of buys back and other information.

In some embodiments of the present invention, obtaining the virtual article purchasing behavior parameter information of the target user may be implemented by:

acquiring original data sets of different dimensions corresponding to the target user; determining a purchase sequence and corresponding purchase time of the virtual article according to the original data sets with different dimensions; and determining the purchasing behavior parameters of the target user virtual article in different unit time based on the purchasing sequence and the corresponding purchasing time of the pseudo article. The behavior of various types of virtual articles purchased by the user matched with the corresponding client can be collected through different program components, the used purchase record log is effectively extracted, and the purchase sequence and the corresponding purchase time of the virtual articles are obtained, for example, the virtual article A, the virtual article B and the virtual article C are purchased in sequence on each day of the week. So in the present application, the triplet information (device number (user account number), purchase time, and purchase sequence) may be extracted to determine the parameters of the virtual article purchase behavior of the target user. In the process, the behaviors of target users who do not successfully purchase the virtual articles can be eliminated by denoising the purchase sequence, and the influence on the accuracy of the virtual article recommendation model due to inaccurate purchase behavior parameters is avoided.

Step 302: and the virtual article recommending device determines a sequence subgraph matched with the virtual article purchasing behavior of the target user based on the virtual article purchasing behavior parameter information of the target user.

The sequence subgraph matched with the virtual article purchasing behavior of the target user can be composed of a formal purchasing sequence in a real item task and a pseudo item sequence in a pseudo item task, specifically: the purpose of constructing the pseudo item recommendation model is to improve the accuracy of the original prediction task and enhance the generalization capability of the virtual item recommendation model by constructing a simpler prediction task (for example, in a use scene recommended by game items, a task for predicting the item type to be purchased by a game player), combining the original task (namely, the purchase task for actually purchasing the item by the player in the game), processing through a neural network model, and sharing multi-task learning modes such as item characteristics, user characteristics, sequence segment characteristics and the like, so that the pseudo item recommendation model is suitable for different use scenes.

Step 303: and the virtual article recommending device determines a graph neural network model in the virtual article recommending model through the sequence subgraph.

Step 304: and the virtual article recommending device determines corresponding sequence segment feature vectors based on a graph neural network model in the virtual article recommending model.

In some embodiments of the present invention, determining the corresponding sequence segment feature vector based on the graph neural network model in the virtual article recommendation model may be implemented by:

linearly weighting all feature nodes of a sequence subgraph in the graph neural network model to determine a first sequence segment feature vector, wherein the first sequence segment feature vector is used for representing the purchasing behavior preference of the target user; extracting the purchase parameters of the last item of the target user from the graph neural network model, and determining a second sequence segment feature vector; and splicing the first sequence segment feature vector and the second sequence segment feature vector to determine a corresponding sequence segment feature vector. The method comprises the steps that the content of a node of a sequence subgraph in a graph neural network model and the content of each type of edge represented are different, a sequence segment feature vector is composed of two parts, one part represents the user preference of the whole sequence and is used for carrying out linear weighting on the feature node of the whole sequence and is called as the first part feature of the sequence; the other part is the user's recent preference, represented by the result of the last purchase, called the second part of the sequence. And splicing the first part of features and the last purchased result to form an integral fragment feature vector of the complete sequence.

Step 305: and the virtual article recommendation device adjusts the network parameters of the virtual article recommendation model based on the sequence segment feature vector and the user feature vector of the target user.

Thereby, matching of the virtual item recommendation model with the target user may be achieved.

The obtaining of the user feature vector can be realized by the following steps:

initializing target user vectors shared by different sequences in the sequence subgraphs matched with the virtual article purchasing behaviors of the target users; processing the target user vector through a first user neural network model in the virtual article recommendation model, and determining a first user single sequence feature matched with a real purchase sequence; processing the target user vector through a second user neural network model in the virtual article recommendation model, and determining a second user single-sequence feature matched with the pseudo item sequence; and splicing the first user single sequence feature and the second user single sequence feature to form a user feature vector of the target user.

Specifically, a user vector of a target user represents common preference of the user to two tasks (respectively corresponding to two different purchase sequences), the user vector is processed by two independent first user neural network models and second user neural network models which have the same size and do not share parameters, the obtained first user single-sequence features and second user single-sequence features represent preference degrees of the user to different characters (namely which virtual article and which type of virtual article the target user prefers), and the user feature vector of the target user can represent preference of the target user in a balanced manner through splicing processing.

In some embodiments of the present invention, based on the sequence segment feature vector and the user feature vector of the target user, adjusting a network parameter of the virtual item recommendation model to match the virtual item recommendation model with the target user may be implemented by:

determining a multitask loss function matched with the virtual article recommendation model; adjusting network parameters of the virtual article recommendation model based on the sequence segment feature vector, the user feature vector of the target user and the multitask loss function; until the loss functions of different dimensions corresponding to the virtual article recommendation model reach corresponding convergence conditions; so as to determine the name of the virtual item purchased again by the target user and the category of the virtual item to be purchased through the virtual item recommendation model.

After training of the virtual item recommendation model is completed, the virtual item recommendation model may be deployed in a server corresponding to a use scenario, and a virtual item recommendation method provided in an embodiment of the present invention is described with reference to fig. 2, fig. 4 is a schematic view of an optional flow of the virtual item recommendation method provided in the embodiment of the present invention, and it can be understood that the steps shown in fig. 4 may be executed by various electronic devices operating the virtual item recommendation device, for example, a dedicated terminal, a server, or a server cluster with the virtual item recommendation device may be used, where the dedicated terminal with the virtual item recommendation device may be the electronic device with the virtual item recommendation device in the embodiment shown in fig. 2 of the foregoing sequence. The following is a description of the steps shown in fig. 4.

Step 401: and responding to the virtual article purchase request, and acquiring the virtual article purchase behavior parameter information of the target user.

when the virtual object is a game item, determining a game role held by the target user, wherein different game roles are matched with the corresponding game item; determining a purchase sequence and a corresponding purchase time of a corresponding game item based on the game character; and determining the virtual article purchasing behavior parameters of the target user in different unit time based on the purchasing sequence and the corresponding purchasing time of the game props. In this case, for example, an RPG (Role-playing game) is assumed to have seven predetermined game roles: darts, catchers, hunters, killers, music, swords, and soldiers, each identity corresponding to a different game property to which a player may choose to join as a target object. The game props as the virtual objects are game props which attack by launching bullets in a virtual environment, or virtual archers and virtual slingshots which launch clusters of arches, and target users can purchase different game props in the virtual environment through different game characters held by the target users and attack through the purchased game props.

Taking a shooting game as an example, a user may control different game characters to freely fall, glide or open a parachute to fall in the sky in the virtual scene through the purchased game items, run, jump, crawl, bow to move ahead on land, or control different game characters to swim, float, or dive in the sea, or the like, and certainly, the user may also control different game characters to ride the game items to move in the virtual scene, for example, the game items may be virtual cars, virtual aircrafts, virtual yachts, and the like, which is only exemplified in the above scene, but is not specifically limited in the embodiment of the present invention. The user can also control different game roles to interact with other different game roles in modes of fighting and the like through the game props, the game props can be cold weapons or hot weapons, and the types of the game props are not particularly limited.

The method provided by the invention can be applied to the recommendation process of game props in different game environments such as a virtual reality application program, a three-dimensional map program, a military simulation program, a First-person shooter game (FPS), a Multiplayer Online Battle Arena game (MOBA), and the like, and the following embodiment is exemplified by the application in the game.

The game based on the virtual environment is often composed of one or more maps of game worlds, the virtual environment in the game simulates the scene of the real world, the user can control different game characters in the game to walk, run, jump, shoot, fight, drive, switch to use the game props, use the game props to attack other different game characters and other actions in the virtual environment, the interactivity is strong, and a plurality of users can form a team on line to play a competitive game. When a user controls different game roles to attack different game roles by using the game props, the user selects proper game props to attack different game roles according to the positions of the different game roles or the operation habits.

Step 402: and determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user.

Step 403: and determining a real purchasing sequence matched with the target user based on the virtual article purchasing behavior parameter information of the target user, and determining a real purchasing sequence of the target user and a sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchasing sequence of the target user.

Step 404: and determining a sequence segment feature vector corresponding to the target user through the sequence subgraph.

In some embodiments of the present invention, determining a sequence subgraph matching with the virtual article purchasing behavior of the target user based on the parameter information of the virtual article purchasing behavior of the target user can be implemented by:

determining a real purchase sequence and a pseudo item sequence based on the virtual item purchase behavior parameter information of the target user; determining nodes of the sequence subgraph by determining nodes of a target virtual article in the real purchase sequence and nodes in a pseudo prop sequence; determining a first type edge of the sequence subgraph based on a chronological order of the target virtual item in the real purchase sequence; determining a second type edge of the sequence subgraph based on the time sequence of the target virtual article in the pseudo prop sequence;

determining a third type edge of the sequence subgraph based on the incidence relation between the real purchase sequence and the pseudo prop sequence; and determining a sequence subgraph matched with the virtual article purchasing behavior of the target user according to the nodes of the sequence subgraph, the first type edge, the second type edge and the third type edge.

In some embodiments of the present invention, the Graph neural network may specifically be a Graph convolution neural network (GCN), and may also be Graph neural networks such as Graph-RNN and Graph-MPNN. Specifically, taking the GCN model framework as an example, the input is a graph, which is transformed by layer-by-layer calculation, and finally a graph is output. The GCN model has the following three properties of deep learning: 1. hierarchical structure, features are extracted layer by layer, and one layer is more abstract and higher level than the other layer. 2. And nonlinear transformation is adopted to increase the expression capability of the model. 3. And (3) end-to-end training, any rule does not need to be defined, and only one mark is needed to be given to the nodes of the graph, so that the model can learn by itself and feature information and structural information are fused.

Step 405: and sharing the sequence segment feature vector corresponding to the target user through a self-attention mechanics learning mechanism of the virtual article recommendation model to form a new sequence segment feature vector.

In some embodiments of the present invention, the sequence segment feature vector is shared by a self-attention learning mechanism of the virtual article recommendation model to form a new sequence segment feature vector, which may be implemented as follows:

initializing target user vectors shared by different sequences in the sequence subgraphs matched with the virtual article purchasing behaviors of the target users; processing the target user vector through a first user neural network model in the virtual article recommendation model, and determining a first user single sequence feature matched with a real purchase sequence; processing the target user vector through a second user neural network model in the virtual article recommendation model, and determining a second user single sequence feature matched with a pseudo item sequence; and splicing the first user single sequence feature and the second user single sequence feature to form a user feature vector of the target user. Taking a property recommendation scene in a WeChat small program game as an example, the self-attention mechanics learning mechanism carries out sharing processing on the sequence segment feature vector of the target user to form a new sequence segment feature vector, so that the new sequence segment feature vector can share the features obtained by the multitask learning mechanism aiming at the property feature, the user feature, the sequence segment feature and the like of the target user, the accuracy of the game property to be recommended to the user is improved, the generalization capability of the virtual article recommendation model is enhanced, and the virtual article recommendation model is suitable for different WeChat small program game use scenes.

Step 406: and determining the virtual articles to be purchased by the target user through the virtual article recommendation model based on the user feature vector of the target user.

The determined virtual item to be purchased by the target user comprises the name of the virtual item to be purchased and the category of the virtual item to be purchased.

When the number of virtual articles is large, sorting recall sequences of the virtual articles based on names of the virtual articles purchased by the target user and categories of the virtual articles to be purchased; and adjusting the sequence of the virtual articles displayed in the user interface according to the sequencing result of the recall sequence of the virtual articles.

In some embodiments of the present invention, a user interface may also be displayed, where the user interface includes a person-name view screen for observing the virtual environment at a person-name view of a different game role, the different game role holds the game prop, and the user interface further includes an injury control component; and controlling the game prop to attack a corresponding injury object in the display user interface through the injury control component so as to realize that different game roles in the display user interface attack the injury object.

In some embodiments of the present invention, when the target user selects any one of the displayed virtual items, a payment method matching the target user may be further determined based on the user feature vector of the target user, or a payment method matching the target user may be further determined based on the user feature vector of the target user and the type information of the virtual item; and triggering a corresponding payment process based on the payment mode matched with the target user.

The virtual article recommendation method provided by the present invention is described below by taking a virtual article as a game item in a game process as an example, where fig. 5 is a schematic view of a use scenario of the virtual article recommendation method provided by the present invention, and referring to fig. 5, a terminal (including a terminal 50-1 and a terminal 50-2) is provided with a client of a same game, and is operated by different users through respective game characters held by the different users, and the users can purchase and use different game items in a game environment through corresponding clients; the terminal is connected to the server 200 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two, and uses a wireless link to realize data transmission.

Referring to fig. 6 and 7, fig. 6 is a schematic diagram of a front end display of a virtual article according to an embodiment of the present invention, and fig. 7 is a schematic diagram of a front end display of a virtual article according to an embodiment of the present invention, in the prior art, a game user usually purchases a virtual article (e.g., a game prop) of interest in a virtual environment, where the prop can be a movable object (game weapon or transportation tool) for any decoration and arrangement in a scene. There are also items in the game that are convenient items for the game target user, which are usually acquired by the game target user for a task or by purchase.

Different game target users often have different prop requirements at different times, and the required props are accurately recommended to the game target users, so that the purchase rate of the game props by the game target users can be improved, the game revenue of game application merchants can be increased, the occurrence of negative emotion of the game target users on game application caused by inaccurate prop recommendation can be reduced, and the stickiness of the game target users on the game application is improved. In order to be able to recommend game items to a user, a server that typically recommends hot game items or newly-placed game items to the user. However, the interesting game items of each user are not the same, and the recommendation of popular items or new shelving items is not suitable for each user, so that the requirement of the accuracy of the recommendation for different users cannot be met.

In the prior art, a game item recommendation method is combined with target user winning rate, purchase data, behavior data and the like for analysis, and a single recommendation model is constructed. The method has the defects that the time series relation of the purchasing behaviors of the user cannot be considered, and the user preference cannot be quickly reflected; meanwhile, the regularity description characteristics of the prop are not mined, and the accurate recommendation cannot be realized only by considering the user behavior singly.

In order to solve the above-mentioned drawbacks, a virtual article recommendation model needs to be trained to implement prejudgment on game items to be purchased by a target user, and a game item serving as a virtual article is timely and accurately recommended to the target user, referring to fig. 8 and 9, where fig. 8 is an optional flow diagram of a training method of the virtual article recommendation model provided in the embodiment of the present invention; fig. 9 is a schematic diagram of data flow in a training process of a virtual item recommendation model according to an embodiment of the present invention, where a virtual item recommendation method according to an embodiment of the present invention includes the following steps:

step 801: and acquiring a prop purchasing sequence of a target user and the category of each prop to form a pseudo prop task.

Taking the first type game as an example, if the actual prop purchase sequence a in a certain week is (silver breaking, gold stone needle converted into a coupon, universal heart law incomplete page (purple)), and the next actual prop purchase sequence is universal heart law incomplete page (purple), the forged prop sequence B is: (silver cullet-basis resource, gold stone needle converted into coupon-stone needle system, general heart law incomplete page (purple) -heart law system prop), the next prediction is classified as a heart law system prop.

Wherein, the purpose of constructing the pseudo prop task is as follows: by constructing a simpler prediction task (in this example, a property category prediction task), learning is performed together with the original task (i.e. a property purchasing task), and by sharing multi-task learning technologies such as property features, user features, sequence segment features and the like, the accuracy of the original prediction task is improved

Step 802: and constructing subgraphs of the real purchase sequence and the pseudo prop sequence by using each single-week data.

In which, with reference to figure 10,fig. 10 is a schematic diagram of a sequence segment subgraph in a training process of a virtual item recommendation model according to an embodiment of the present invention; wherein there are two types of feature nodes in the graph, including feature vector er _, with dimensions dr and df _i And ef \ u _i Respectively representing the node of the prop i in the real purchase sequence and the node of the pseudo prop sequence. In fig. 10, there are three types of edges (shown by different legends) in total, and the first type of edges are connected in time order of the item purchase sequence, such as (er _ i) ₁ ，er_i ₂ ) (ii) a The second category of edges is the task of forging props, such as (ef _ i) ₁ ，ef_i ₂ ) (ii) a The third class of edges is the associated edge between two tasks, the packet (er _ i) ₁ ，ef_i ₂ ) And (ef _ i) ₁ ，er_i ₂ ) Corresponding edge weights are each W _rr ， _Wff ，W _rf And W _fr . The three types of edges represent the interaction relationship between different types of sequence nodes, so different update weights W are set _rr ，W _ff ，W _rf And W _fr . So that the following figures are given as examples,

ef_ _j ＝W _ff * _{ef_(j-1)} +W _rf *er_ _(j-1) ，

_er _j＝W _rr *er_ _(j-1) +W _fr * _{ef_(j-1)} ，j＝i ₁ ，i ₂ ，…，i ₅ 。

step 803: after a subgraph is formed, two sequence segment feature vectors sr and sf with the length ds are determined by constructing a matched graph neural network.

Specifically, graph Neural Network (GNN): a neural network directly acting on a graph structure mainly processes data of a non-Euclidean space structure (graph structure). Have an input order that ignores nodes; in the calculation process, the representation of the node is influenced by the surrounding neighbor nodes, and the connection of the graph is unchanged; the representation of graph structure enables graph-based reasoning. In general, a graph neural network consists of two modules: the system comprises a Propagation Module (Propagation Module) and an Output Module (Output Module), wherein the Propagation Module is used for transmitting information between nodes in the graph and updating the state, and the Output Module is used for defining an objective function according to different tasks based on vector representation of the nodes and edges of the graph. The graph neural network has: graph Convolutional Neural Networks (GCNs), gated Graph Neural Networks (GGNNs), and Graph attention Neural Networks (GAT) based on attention mechanism.

The sequence segment feature vector is composed of two parts, one part represents the user preference of the whole sequence, and the first part of the feature sr1= ∑ Σ of the sequence is called for linearly weighting the feature nodes of the whole sequence _{j＝i1......i5α} rj er _ j, where α rj is a parameter. The other part is the recent preferences of the user, denoted by the last bought result sr2= er _ i5, referred to as the second part of the sequence. Splicing the first partial feature sr1 and the last purchased result sr2 to form an overall segment feature vector sr = [ sr1, sr2 ] of the sequence]. The same principle can be used to obtain sr = [ sf1, sf 2=]Where sf1= ∑ Σ _{j＝i1......iα} f _j *er_j，sf2＝ef_i5，αf _j Are parameters.

Step 804: and sharing the sr and the sf in a self-attention learning mode.

Referring to fig. 11, fig. 11 is a schematic diagram of a sequence segment self-attention mechanics learning process in a training process of a virtual item recommendation model according to an embodiment of the present invention; the self-attention learning mode comprises the following steps:

1)s11＝sr*sr，s12＝sr*sf；s21＝sf*sr，s22＝sf*sf；

2)s11’＝e^s11/(e^s11+e^s12)，s12’＝e^s21/(e^s11+e^s12)；

s21’＝e^s21/(e^s21+e^s22)，s22’＝e^s22/(e^s21+e^s22)，

3)sr’＝s11’*sr+s12’*sf；sf’＝s21’*sr+s22’*sf。

sharing sr and sf in a self-attention learning mode as follows to obtain new session state sr 'and sf'. Thereby, a better sharing of features between the two tasks may be enabled.

Step 805: the corresponding user features ur 'and uf' are learned.

Specifically, referring to fig. 12, fig. 12 is a schematic diagram illustrating prediction performed by sequence segments and user features in a training process of a virtual article recommendation model according to an embodiment of the present invention, where the user features are composed of two parts, a first part represents a shared user feature us of two sequences, and another part is a single-sequence user feature uf and ur. Specifically, it can be expressed as follows:

1) The initialization sequence shares a user feature vector us, which represents the user's common preferences for the two tasks

2) The us passes through two independent neural network layers with the same size and unshared parameters, so that ur and uf are obtained, and represent the individual preference of the user for the two tasks

3) And respectively connecting us, ur and uf (directly splicing into a longer characteristic vector), and forming new user characteristics ur '= [ ur, us ] and uf' = [ uf, us ], so that a comprehensive characteristic vector which considers the common preference of the user and the personal washing of the user can be obtained.

Step 806: and respectively predicting the game items to be purchased by the user by using the vectors [ sr ', ur ' ] and [ sf, sr ' ] obtained by connection.

Wherein, the predicted game item to be purchased by the user may include: and (3) purchasing the props ID (actual purchasing forecast task) at the next moment and purchasing prop categories (pseudo props task) by the user at the next moment, so that training of a graph neural network for recommending game props can be completed.

Furthermore, the overall multitask loss can be obtained through linear weighting with a certain weight, the loss gradient is subjected to combined learning, the propagation of the inverse gradient is carried out, and er _ i, ef _ i and us are updated.

Specifically, the method comprises the following steps: firstly, linear weighting processing is carried out, which comprises the following steps: for two classification tasks, positive samples of the classification tasks are respectively the actual next purchased prop and the corresponding prop category, and the rest are negative samples. The cross entropy of softmax can be obtained by using the results obtained from the real sample and the prediction layer as loss, which is loss _ r and loss _ f, respectively, and then the overall loss function loss can be expressed as:

loss＝αr ₁ *loss_r+αf ₁ *loss_f，

wherein, α r ₁ And α f ₁ Are all parameters. Then entering a subsequent joint learning phase: directly solving the gradient of loss as a simple optimization problem, and finally carrying out inverse gradient propagation. Certainly, the data quantity can be further increased through multi-week data and multi-time single-week data after multi-time data enhancement, the result accuracy is improved, and after training is completed, game prop recommendation is performed through the virtual article recommendation model provided by the invention, and the method comprises the following steps:

1) Acquiring a prop purchasing sequence of a target player and the category of each prop to form a pseudo prop task; if a real prop in a certain week has a purchase sequence A (silver breaking, gold stone needle converted into a coupon, general heart law incomplete page (purple)), an operator does not know that the next real prop to be purchased is the general heart law incomplete page (purple). The forged prop sequence B is as follows: (silver breaking-basic resources, golden stone needle converted into coupon-stone needle system and general heart law incomplete page (purple) -heart law system prop), and the next prediction classification real result is unknown as the heart law system prop.

2) Each single week (as a unit of time, users with frequent triggers to game applets may also

Obtaining purchase data through daily user logs) data structure real purchase sequence and subgraph of pseudo prop sequence, and obtaining ef _, obtained by training before through table look-up _j And er _j ，j＝i1，i2，…，i5。

3) After a subgraph is constructed, sequence segment feature vectors (session state) sr and sf are obtained. The sequence segment feature vector is composed of two parts, one part represents the user preference of the whole sequence, and in order to linearly weight the feature nodes of the whole sequence, the first part of the feature of the sequence is called,

wherein α r _j Is a parameter. Another part is the recent preference of the user, using the last purchased result sr2= er _ i5 denotes the second partial feature called sequence. Splicing the first partial feature sr1 and the last purchased result sr2 to form an overall segment feature vector sr = [ sr1, sr2 ] of the sequence]. The same principle can be used to obtain sr = [ sf1, sf 2=]In which>

sf ₂ ＝ef_i5，αf _j Are parameters.

4) Sharing sr and sf through the following self-attention learning method to obtain new session state sr 'and sf', specifically including:

a)s11＝sr*sr，s12＝sr*sf；s21＝sf*sr，s22＝sf*sf；

b)s11’＝e^s11/(e^s11+e^s12)；s12’＝e^s21/(e^s11+e^s12)；s21’＝e^s21/(e^s21+e^s22)，s22’＝e^s22/(e^s21+e^s22)；

c)sr’＝s11’*sr+s12’*sf；sf’＝s21’*sr+s22’*sf。

the game props of different target users have different use habits, and game roles, skills and weapons used in a game environment are different, so that sequence segment feature vectors of the target users are shared through a self-attention mechanical learning mechanism, and weights of different feature vectors are adjusted through the attention mechanical learning mechanism when the types and names of the game props to be purchased by the users are predicted to form new sequence segment feature vectors, so that the new sequence segment feature vectors can share features obtained by a multi-task learning mechanism aiming at prop features, user features, sequence segment features and the like of the target users, the accuracy of the game props to be recommended to the users is improved, the generalization capability of a virtual article recommendation model is enhanced, the virtual article recommendation model is adaptive to different WeChat program game use scenes, and meanwhile, parameters of a graph neural network based on the attention mechanical system in the virtual article recommendation model can be adjusted in a training stage, the virtual article model is rapidly deployed, and the waiting time of the users is saved.

5) And looking up a table to obtain corresponding user characteristics ur 'and uf'.

6) And respectively predicting the purchase item ID (real purchase prediction task) of the user at the next moment and the purchase item category (pseudo item task) of the user at the next moment by using the vectors [ sr ', ur' ] and [ sf ', sr' ] obtained by connection. Therefore, the virtual article can be accurately recommended to the user by using the virtual article recommendation model, the time for the user to search the virtual article is reduced, and the use experience of the user is improved.

Next, the following description will be made by taking a single-week purchase sequence of a certain user (user H) in a first type game process as an example, where the game item recommendation process provided by the present invention is determined, where the actual one-week item purchase sequence a of the user H is: silver breaking, golden stone needle coupon, universal heart method waste pages (purple), and the next real purchased prop is the universal heart method waste pages (purple) as a classified real result. Then the first sequence of forged props is constructed as follows: (silver breaking-basic resources, gold stone needles converted into coupons-stone needles system, general heart law residual page (purple) -heart law system prop), and the next prediction classification result is the heart law system prop.

Referring to fig. 13, fig. 13 is a schematic view of processing sequence segments in a process of recommending game items by using a virtual item recommendation model according to an embodiment of the present invention, where a process of constructing a subgraph includes: wherein er1 represents the feature vector of silver scrap, er2 represents the feature vector of the golden stone needle conversion coupon, and er3 represents the feature vector of a general heart method incomplete page (purple); ef1 represents the feature vector of the silver scrap-basic resource, ef2 represents the feature vector of the golden stone needle converted into the coupon-stone needle system, and ef3 represents the feature vector of the general cardiac method incomplete page (purple) -cardiac method system prop. Simultaneously, initialize er1, er2, er3, ef1, ef2, ef3, specifically:

1)ef1’＝W _ff *ef1+W _rf *er1；

2)er1’＝W _rr *er1+W _fr *ef1，

3)ef2’＝W _ff *ef1+W _rf *er1；

4)er2’＝W _rr *er1+W _fr *ef1；

5)ef3’＝W _ff *ef2+W _rf * er2; ander3＝W _rr *er2+W _fr * ef2, thereby obtaining a sequence segment feature vector, namely:

sr＝αr ₁ *er ₁ +αr ₂ *er ₁ +αr ₃ *er ₂ +αr ₄ *er ₃

sf＝αf ₁ *ef ₁ +αf ₂ *ef ₁ +αf ₃ *ef ₂ +αf ₄ *ef ₃ 。

and then, adjusting the weights of different characteristic vectors through a self-attention mode mechanism for processing to obtain new characteristic vectors sr 'and sf'.

And the user characteristics of the target user H are ur '= [ Wur us, us ] and uf' = [ Wuf us, us ], and the user characteristics are substituted into the sequence segment characteristic vector to obtain final predicted layer input vectors [ sr ', ur' ] and [ sf ', uf' ]. The real labels of the two prediction tasks are general heart law residual page (purple) and heart law system property, errors loss _ r and loss _ f of the prediction result and the real result are obtained, and the overall loss function loss obtained by linear weighting is as follows:

loss＝αr ₁ *loss_r+αf ₁ *loss_f

and then carrying out gradient descent back propagation processing to update all variables. And determining corresponding training samples by using the purchase record information in the game system to train the virtual article recommendation model, and recommending the target user H by a certain week operator. Determining a week property purchase sequence A on a user target user H through a purchase record in a game server: silver breaking, general heart method incomplete page (purple), and constructing a corresponding forged prop sequence B: silver breaking-basic resource, general heart law incomplete page (purple) -heart law system prop. The lookup table knows their feature vectors are er _ sui, er _ xinzi, ef _ sui and ef _ xinzi, respectively, assuming α r ₁ All are 0.5, then the corresponding sequence fragment feature vectors are [0.5 der _suri +0.5 der _xinzi, er _xinzi]And [0.5 × ef _sui +0.5 × ef _xinzi, ef _xinzi]And then the conversion is carried out in a self-attention mode of the preamble to obtain sr 'and sf'.

Meanwhile, table lookup is carried out to obtain user characteristics of the target user H, namely ur _ hong and uf _ hong, so that [ sr ', ur _ hong ] and [ sf', uf _ hong ] can be used as input vectors and input into a prediction layer to obtain a prediction probability result of the global game prop. Therefore, the accuracy of the game props to be recommended to the user can be improved, the universal heart law incomplete pages (purple) in the heart law system prop types are accurately recommended to the user to adapt to the subsequent game control of the game roles controlled by the user, and the target user can conveniently search the prop stores for a long time to obtain the required game props.

Fig. 14 is a schematic diagram of a display of the front end of a virtual article adjusted by a display sequence according to an embodiment of the present invention, in which a recall sequence of a game item recommended to a user is sorted by a result of a predicted probability of the game item in the global sense; and adjusting the sequence of the game props displayed in the user interface according to the sequencing result of the recall sequence of the game props.

And finally, according to the size of the page, the prop with the highest probability is selected and displayed on the game page, or the sequence of the game props to be displayed is adjusted so as to better meet the purchase intention of the user.

The beneficial technical effects are as follows:

compared with the prior art, the technical scheme provided by the application can be used for obtaining the virtual article purchasing behavior parameter information of the target user; determining a sequence subgraph matched with the virtual article purchasing behavior of the target user; determining a graph neural network model in the virtual article recommendation model through the sequence subgraphs; determining corresponding sequence segment feature vectors based on a graph neural network model in the virtual article recommendation model; based on the sequence segment feature vector and the user feature vector of the target user, network parameters of the virtual article recommendation model are adjusted, the virtual article recommendation model is matched with the target user, virtual articles can be accurately recommended to the user by the virtual article recommendation model, time for the user to search for the virtual articles is reduced, and use experience of the user is improved. With reference to tables 1 to 3, when the virtual article recommendation model provided by the invention is used, it can be determined that the game props recommended by the virtual article recommendation method provided by the invention better meet the purchase will of the user by comparing different dimensions with the conventional technology, and the viscosity of the user is improved. The different dimensions include respective corresponding evaluation indexes, namely, hit rate (Hit ratio), matching degree MRR (Mean correct Rank), and Normalized broken cumulative gain NDCG (Normalized broken cumulative gain).

TABLE 1

TABLE 2

TABLE 3

The above description is only exemplary of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for virtual item recommendation, the method comprising:

acquiring virtual article purchasing behavior parameter information of a target user;

determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the virtual item purchasing behavior parameter information of the target user, wherein the pseudo item task is a counterfeiting prediction task of a known result according to game planning professional knowledge;

initializing target user vectors shared by different sequences in the sequence subgraphs matched with the virtual article purchasing behaviors of the target users;

processing the target user vector through a first user neural network model in a virtual article recommendation model, and determining a first user single sequence feature matched with a real purchase sequence;

processing the target user vector through a second user neural network model in the virtual article recommendation model, and determining a second user single sequence feature matched with a pseudo item sequence;

splicing the first user single-sequence feature and the second user single-sequence feature to form a user feature vector of the target user;

2. The method according to claim 1, wherein said determining a corresponding sequence segment feature vector based on said pseudo item task and said target user's true purchase sequence comprises:

determining a real purchase sequence matched with the target user based on the virtual article purchase behavior parameter information of the target user;

determining a real purchase sequence of the target user and a sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchase sequence of the target user;

and determining a sequence segment feature vector corresponding to the target user through the sequence subgraph.

3. The method according to claim 2, wherein the determining the real purchase sequence of the target user and the sequence subgraph corresponding to the pseudo item sequence based on the pseudo item task and the real purchase sequence of the target user comprises:

determining a real purchase sequence and a pseudo item sequence based on the virtual item purchase behavior parameter information of the target user;

determining nodes of the sequence subgraph by determining nodes of a target virtual article in the real purchase sequence and nodes in a pseudo prop sequence;

determining a first type edge of the sequence subgraph based on a chronological order of the target virtual item in the real purchase sequence;

determining a second type edge of the sequence subgraph based on the time sequence of the target virtual article in the pseudo prop sequence;

determining a third type edge of the sequence subgraph based on the incidence relation between the real purchase sequence and the pseudo-item sequence;

and determining a sequence subgraph matched with the virtual article purchasing behavior of the target user according to the nodes of the sequence subgraph, the first type edge, the second type edge and the third type edge.

4. The method according to claim 1, wherein said determining a corresponding sequence segment feature vector based on said pseudo item task and said target user's true purchase sequence comprises:

linearly weighting all feature nodes of the sequence subgraph to determine a first sequence segment feature vector, wherein the first sequence segment feature vector is used for representing the purchasing behavior preference of the target user;

extracting purchase parameters of the last simulated object of the target user, and determining a second sequence segment feature vector;

and splicing the first sequence segment feature vector and the second sequence segment feature vector to determine a corresponding sequence segment feature vector.

5. The method of claim 1, further comprising:

determining a multitask loss function matched with the virtual article recommendation model;

adjusting network parameters of the virtual article recommendation model based on the sequence segment feature vectors, the user feature vectors of the target users and the multitask loss function;

until loss functions of different dimensions corresponding to the virtual article recommendation model reach corresponding convergence conditions; so as to realize the matching of the virtual item recommendation model and the target user.

6. The method of claim 1, further comprising:

determining the name of the virtual article to be purchased by the target user and the category of the virtual article to be purchased;

sorting recall sequences of the virtual items based on the names of the virtual items purchased by the target users and the categories of the virtual items to be purchased;

and adjusting the sequence of the virtual articles displayed in the user interface according to the sequencing result of the recall sequence of the virtual articles.

7. The method of claim 1, wherein the obtaining of the virtual article purchasing behavior parameter information of the target user comprises:

when the virtual object is a game item, determining a game role held by the target user, wherein different game roles are matched with the corresponding game item;

determining a purchase sequence and a corresponding purchase time of a corresponding game item based on the game character;

and determining the virtual article purchasing behavior parameters of the target user in different unit time based on the purchasing sequence and the corresponding purchasing time of the game props.

8. The method of claim 7, further comprising:

displaying a user interface, wherein the user interface comprises a personal perspective view picture for observing a virtual environment by using personal perspective views of different game roles, the different game roles hold the game props, and the user interface further comprises an injury control component;

and controlling the game prop to attack a corresponding injury object in the display user interface through the injury control component so as to realize that different game roles in the display user interface attack the injury object.

9. The method of claim 1, further comprising:

when the target user selects any of the presented virtual items,

determining a payment mode matched with the target user based on the user feature vector of the target user, or,

determining a payment mode matched with the target user based on the user feature vector of the target user and the type information of the virtual article;

and triggering a corresponding payment process based on the payment mode matched with the target user.

10. A virtual item recommendation apparatus, characterized in that the apparatus comprises:

the information transmission module is used for acquiring the virtual article purchasing behavior parameter information of the target user;

the information processing module is used for determining a pseudo item task matched with the virtual item purchasing behavior of the target user based on the parameter information of the virtual item purchasing behavior of the target user, wherein the pseudo item task is a counterfeiting prediction task with a known result, which is carried out according to game planning professional knowledge;

the information processing module is used for processing the target user vector through a first user neural network model in a virtual article recommendation model and determining a first user single sequence feature matched with a real purchase sequence;

the information processing module is used for processing the target user vector through a second user neural network model in the virtual article recommendation model and determining a second user single sequence feature matched with the pseudo item sequence;

the information processing module is used for splicing the first user single sequence feature and the second user single sequence feature to form a user feature vector of the target user;

the information processing module is used for determining the virtual articles to be purchased by the target user based on the user feature vector of the target user.

11. The apparatus of claim 10,

the information transmission module is used for determining game roles held by the target user when the virtual object is a game prop, wherein different game roles are matched with the corresponding game props;

the information transmission module is used for determining a purchase sequence and a corresponding purchase time of the corresponding game prop based on the game role;

and the information transmission module is used for determining the purchasing behavior parameters of the virtual object of the target user in different unit time based on the purchasing sequence and the corresponding purchasing time of the game props.

12. The apparatus of claim 10,

the information processing module is used for displaying a user interface, the user interface comprises a person name visual angle picture for observing a virtual environment by using person name visual angles of different game roles, the different game roles hold game props, and the user interface also comprises an injury control component;

the information processing module is used for controlling the game prop to attack a corresponding injury object in the display user interface through the injury control component so as to realize that different game roles in the display user interface attack the injury object.

13. An electronic device, characterized in that the electronic device comprises:

a memory for storing executable instructions;

a processor configured to execute the executable instructions stored in the memory to implement the virtual item recommendation method of any one of claims 1 to 9.

14. A computer-readable storage medium storing executable instructions, wherein the executable instructions when executed by a processor implement the virtual item recommendation method of any one of claims 1-9.