A method and device for predicting user behavior in a metaverse space, and a storage medium

Abstract

This invention discloses a method, apparatus, and storage medium for predicting user behavior in a metaverse space. The method includes: acquiring a set of behavioral data of users in the metaverse space during a preset time period; processing the behavioral data set into model sample data for multiple time steps; retrieving a pre-trained behavior prediction model, and inputting the corresponding model sample data into the behavior prediction model according to the time steps to obtain a final behavioral feature data; and acquiring the typical behavior corresponding to the final behavioral feature data as the prediction result. Using the embodiments disclosed in this invention, it is possible to predict the actions and behaviors that users in the metaverse space will perform based on recorded data of individual behavior in the metaverse space along the time dimension. This is beneficial for monitoring user behavior in the metaverse space to maintain order in the metaverse space, and can provide data for obtaining user profiles for various commercial applications applied to the metaverse.

Description

Technical Field

[0001] This invention relates to the field of virtual world technology, and in particular to a method, device and storage medium for predicting user behavior in a metaverse space. Background Technology

[0002] The metaverse, also known as the virtual world, is an open and shared online platform that integrates information technology, communication technology, AR, VR, and other virtual technologies. It is a vast and evolving virtual universe. In the metaverse, user behavior is not restricted by system-defined actions, possessing a high degree of freedom. However, as a multi-user shared platform and activity space, the metaverse inevitably has certain behavioral rules to regulate user behavior and maintain order. Therefore, in terms of social governance, it is necessary to analyze user behavior and further, obtain group characteristics to better manage social order. For certain commercial applications targeting metaverse users, it is necessary to analyze user behavior to understand their habits and preferences, thereby creating user profiles. Therefore, it is necessary to monitor users in certain scenarios by predicting their behavior, thereby maintaining order in the metaverse or meeting the needs of certain commercial applications. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method, apparatus, electronic device, and computer-readable storage medium for predicting user behavior in the metaverse space.

[0004] In a first aspect, embodiments of the present invention provide a method for predicting user behavior in a metaverse space, comprising the following steps:

[0005] Obtain a set of behavioral data of a user in the metaverse space for the current preset time period, wherein the set of behavioral data is a time-series set including behavioral data at multiple moments, and the behavioral data at each moment is composed of static action feature data and dynamic performance data of body parts;

[0006] The time-series data set is processed into model sample data at multiple time steps, the number of time steps corresponding to the number of moments in the time-series data set; a pre-trained behavior prediction model is retrieved, and the corresponding model sample data is input into the behavior prediction model according to the time steps to obtain a final behavior feature data; and

[0007] The typical behaviors corresponding to the final behavioral feature data are obtained as the prediction results.

[0008] Secondly, embodiments of the present invention provide a device for predicting user behavior in a metaverse space, comprising: a data acquisition module, a sample processing module, a prediction module, and a behavior recognition module. The data acquisition module acquires a set of behavioral data for a user in the metaverse space during a preset time period. This behavioral data set is a time-series data set including behavioral data at multiple time points, where each time-series behavioral data is composed of static action feature data and dynamic body part performance data. The sample processing module is connected to the data acquisition module and processes the time-series data set into model sample data for multiple time steps. The prediction module is connected to the sample processing module and retrieves a pre-trained behavior prediction model, inputting the corresponding model sample data to the behavior prediction model according to the time steps to obtain a final behavioral feature data. The behavior recognition module is connected to the prediction module and acquires typical behaviors corresponding to the final behavioral feature data as prediction results.

[0009] Thirdly, embodiments of the present invention provide an electronic device, the electronic device comprising: a processor and a memory storing computer program instructions; wherein the processor executes the computer program instructions to implement the steps of the method for predicting user behavior in the metaverse space as described above.

[0010] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps of the method for predicting user behavior in the metaverse space as described above.

[0011] Fifthly, embodiments of the present invention provide a computer program product, the computer program product including computer program instructions, which, when executed by a processor, implement the steps of the method for predicting user behavior in the metaverse space as described above.

[0012] This invention predicts the actions and behaviors that users in the metaverse space will perform based on data from personal behavior records in the metaverse space over time. This is beneficial for monitoring user behavior in the metaverse space to maintain order in the metaverse space. Furthermore, it can provide data for obtaining user profiles for various commercial applications applied to the metaverse, thereby promoting the vigorous and orderly development of the metaverse. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below.

[0014] Figure 1 This is a schematic diagram of the metaverse system structure according to an embodiment of the present invention;

[0015] Figure 2 This is a flowchart of a method for predicting user behavior in the metaverse space according to an embodiment of the present invention;

[0016] Figures 3A-3F These are schematic diagrams of running actions corresponding to different "running" feature datasets of metaverse users according to an embodiment of the present invention.

[0017] Figure 4A This is a schematic diagram of a neuron network unit structure of an LSTM model according to an embodiment of the present invention;

[0018] Figure 4B-4E These are schematic diagrams of the structure of a neuron network unit of an LSTM model according to an embodiment of the present invention during different computation processes.

[0019] Figure 5 This is a flowchart of the training process of an LSTM model according to an embodiment of the present invention;

[0020] Figure 6 A flowchart of a method for obtaining final behavioral feature data based on a behavioral prediction model according to an embodiment of the present invention;

[0021] Figures 7A-7B These are schematic diagrams of the structure of a DRNN model according to an embodiment of the present invention;

[0022] Figure 8 This is a flowchart of a method for obtaining user behavior data in the metaverse space according to an embodiment of the present invention;

[0023] Figure 9 This is a schematic diagram of a data structure according to an embodiment of the present invention;

[0024] Figure 10 This is a schematic diagram of an action for recovering user behavior data based on a metaverse space according to an embodiment of the present invention;

[0025] Figure 11 This is a flowchart of a method for obtaining dynamic performance data of body parts according to an embodiment of the present invention;

[0026] Figure 12 This is a block diagram illustrating the principle of a device for predicting user behavior in the metaverse space according to an embodiment of the present invention.

[0027] Figure 13 This is a block diagram illustrating the principle of the prediction module according to an embodiment of the present invention;

[0028] Figure 14 This is a principle block diagram of a behavior recognition module according to an embodiment of the present invention;

[0029] Figure 15This is a schematic block diagram of a data acquisition module according to an embodiment of the present invention;

[0030] Figure 16 This is a block diagram of a data acquisition unit according to an embodiment of the present invention;

[0031] Figure 17 This is a block diagram of a static motion feature unit according to an embodiment of the present invention;

[0032] Figure 18 This is a block diagram illustrating the principle of a dynamic representation data unit according to an embodiment of the present invention; and

[0033] Figure 19 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0034] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided to make the principles and spirit of the present invention clearer and more thorough, enabling those skilled in the art to better understand and implement the principles and spirit of the present invention. The exemplary embodiments provided herein are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments described herein without inventive effort are within the scope of protection of the present invention.

[0035] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, computer-readable storage medium, or computer program product. Therefore, this disclosure can be specifically implemented in at least one of the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0036] According to embodiments of the present invention, the present invention claims protection for a method, apparatus, electronic device, computer-readable storage medium, and computer program product for acquiring user behavior data in the metaverse space.

[0037] In this document, terms such as "first" and "second" are used only to distinguish one entity (or operation) from another, and are not intended to require or imply any order or relationship between these entities (or operations). In this document, elements defined by the phrase "including..." (e.g., components, parts, processes, steps) do not exclude the existence of other elements besides those listed; that is, other elements not explicitly listed may also be included. In this document, any elements and their quantities in the accompanying drawings are for illustrative purposes only and not for limitation, and any names in the drawings are for distinction only and have no limiting meaning.

[0038] The principles and spirit of the present invention will be explained in detail below with reference to several exemplary or representative embodiments.

[0039] Figure 1 This is a schematic diagram of a metaverse system structure according to an embodiment of the present invention. The metaverse system includes multiple decentralized data processing nodes for generating and maintaining the metaverse space and the computation of various activities within it. Each data processing node is connected to a database, which stores metaverse space data, processing data for maintaining activities within the space, and so on. In one embodiment, any entity in the metaverse space, including a user within it, is represented by a three-dimensional point cloud array and texture image describing the spatial geometry. The three-dimensional point cloud array is a dataset representing the three-dimensional spatial structure of an entity (such as a user's body). In another embodiment, the metaverse system updates the spatial data at a certain refresh rate and records the data after each refresh. Therefore, when a user engages in various activities in the metaverse space, such as walking, standing, or running, the database records the three-dimensional spatial structure dataset of the user's body parts in the metaverse space at each refresh. This invention refers to these three-dimensional spatial structure datasets representing user actions and behaviors as behavioral performance data. The behavioral performance data is a set of data describing the corresponding behaviors when the user engages in these activities; it is a time-series data set.

[0040] Figure 2 This is a flowchart of a method for predicting user behavior in a metaverse space according to an embodiment of the present invention, which includes the following steps:

[0041] Step S1: Obtain the user's behavior data set for the current preset time period in the metaverse space. The behavior data set is a time-series data set including multiple moment-based behavior datasets. The moment-based behavior data consists of static action feature data and dynamic body part performance data. In one embodiment, the metaverse system not only records the three-dimensional spatial structure dataset of the user's body parts in the metaverse space (the three-dimensional spatial structure data being a type of static performance data), but also generates static action feature data and dynamic body part performance data based on the static performance data at each moment. This data is then encoded based on the body parts to obtain a multi-dimensional matrix composed of static action feature data and dynamic body part performance data, denoted as user behavior data X, where X = [S, A], S = [S...]. T1 S T2 ...S Tn ] represents the static motion feature data corresponding to n analysis times, where the static motion feature S at the i-th analysis time is... Ti =[M hi M bi M ali ,……M lri ], Mhi M represents the static head behavior data at the i-th analysis time. bi For the static trunk performance data at the i-th analysis time, M ali M represents the static performance data of the left arm at the i-th analysis time. lri Let A represent the static performance data of the right leg at the i-th analysis time. hd M bd ,……M lrd ] represents the dynamic feature data of all body parts corresponding to n analysis times. M hd Let M be the head dynamic performance data at the nth analysis time, where M hd =[M hd1 M hd2 ...M hd(n-1) ], M hd(n-1) =M hn -M h(n-1) .

[0042] Step S2 involves processing the time-series data set into model sample data for multiple time steps. In one embodiment, the number of time steps corresponds to the number of moments in the time-series data set. For example, if the user's behavior data X includes n moments, then the time-series data set is processed into model sample data for n time steps. For example, the model sample data X1 for the first time step is X1 = [S T1 [A1], S T1 =[M h1 M b1 M al1 ,……M lr1 ], A1 = [M hd1 M bd1 , ...M lrd1 This process is repeated to obtain model sample data for multiple time steps.

[0043] Step S3: Retrieve the pre-trained behavior prediction model and input the corresponding model sample data into the behavior prediction model according to the time step to obtain a final behavior feature data. This invention stores a collection of various typical behavior feature data, such as multiple feature data for "running," multiple feature data for "walking," multiple feature data for "dancing," etc. This invention describes the user's action behavior features using static action feature data and dynamic performance data. The encoding format of the feature data is the same as the data encoding format in the aforementioned time-series data set. In a better embodiment, behaviors have different classifications, and each type of behavior has multiple typical behavior features, corresponding to corresponding feature data. For example, "running" has multiple typical characteristic actions, each with corresponding feature data, such as... Figures 3A-3FThe diagrams shown represent different running action illustrations corresponding to user "running" feature data in the metaverse. This invention pre-trains the behavior prediction model using a dataset of user behavior data from the metaverse space as samples. The model learns typical behavioral features from these samples and uses the feature data corresponding to these behaviors as its output.

[0044] Step S4: Obtain the typical behavior corresponding to the final behavioral feature data as the prediction result. In one embodiment, there is a one-to-one correspondence between the behavior with typical features and its feature data, or one behavior corresponds to multiple feature data, and a corresponding correspondence table is established. By reading the correspondence table, the typical behavior corresponding to the feature data can be determined.

[0045] In one embodiment, the behavior prediction model in step S3 is an LSTM (Long Short-Term Memory) neural network model, or simply the LSTM model. Figure 4A The diagram shown illustrates the structure of a neuron (or "cell") in an LSTM model. The circles represent pointwise operations (calculating the function value at each point in the domain), the boxes represent matrices (i.e., the learned neural network layers), and the lines represent transmitted vectors. t For the sample data at time step t, h t This is the output data calculated by the LSTM model. As can be seen from the network unit structure diagram of the LSTM model, the output data at each time step is also returned to the input as input data, and is input into the model along with the original sample data.

[0046] In LSTM model computation, the first step is to determine the information to discard from the "cell". For example... Figure 4B As shown, the sample data x at the current time step t and the output data h of the previous time step t-1 The input is fed into the first Sigmoid layer, which acts as a forget gate. The information to be discarded is determined by the function relationship shown in Equation 1-1. After the Sigmoid layer operation, if the output result is 1, it means "completely retained"; if the output result is 0, it means "completely discarded".

[0047] Then determine the new information stored in the "cell" state. For example... Figure 4C As shown, to accomplish this function, the "cell" comprises two layers: a second sigmoid layer acts as the "input gate," determining the value i to be updated; a tanh (bipolar) layer creates a new candidate value vector. Added to the state. In this embodiment, new behavioral data is added to the cell state to replace the old behavioral data. The functional relationship of the "input gate layer" is shown in Equation 1-2, and the functional relationship of the tanh (bipolar) layer is shown in Equation 1-3.

[0048] Then update the old "cell" state, which is C. t-1 Updated to C t .like Figure 4D As shown, the specific processing procedure is to change the old state C. t-1 with f t Multiply to discard information that needs to be discarded, then add... Then a new candidate value C was obtained. t Specifically, as shown in Formulas 1-4. The new candidate value varies depending on whether the previously determined discarded information or the stored new information is used, and also varies with the degree to which each state is updated.

[0049] Finally, determine the "cell" output. For example... Figure 4E As shown in the diagram, the third sigmoid layer is run first. Then, the updated "cell" state is processed by the second tanh layer (resulting in a value between -1 and 1). Finally, this value is multiplied by the output of the third sigmoid layer, and only the data that determines the output is ultimately output. The functional relationship is shown in Equations 1-5 and 1-6.

[0050] f t =σ(w f *[h t-1 ,x t ]+b f ) 1-1

[0051] i t =σ(w i *[h t-1 ,x t ]+b i ) 1-2

[0052]

[0053]

[0054] o t =σ(w o *[h t-1 ,x t ]+b o ) 1-5

[0055] h t =o t *tanh(C t ) 1-6

[0056] The prediction model based on LSTM neural networks focuses more on the correlation between different behavioral data over time. It uses the behavioral data from the previous time step as input and the behavioral data state and changes from the next time step as output, and so on, to learn autonomously and gradually pass historical data forward. The "forget gate" of the "cell" controls the internal state information. The gate's input is the output of the hidden node from the previous time step and the current input. The activation function is sigmoid. Because the input of sigmoid is between 0 and 1, multiplying the output of the internal state node with the output of the forget gate controls the amount of information, ensuring that the model can absorb effective information from various influencing factors during historical environmental changes while ignoring invalid and abrupt information. Therefore, the LSTM model can process time-series data and determine the usefulness of input information. Thus, in one embodiment of this invention, an LSTM model is used to predict the behavior of users in the metaverse space after learning from their behavioral data.

[0057] The training process of an LSTM model includes, for example, the following steps: Figure 5 As shown:

[0058] Step S301: Obtain training data. In this embodiment, the training data includes, for example, multiple sets of user behavior data of different types known to users within a given time period. In the following description, unless otherwise specified, the user behavior data refers to the aforementioned data composed of static action feature data and dynamic body part performance data, and has corresponding encoding. Each sample represents a set of user behavior data for a given time period.

[0059] Step S302: Segment the dataset and determine the input and output data. Divide the training data into a training set and a test set, and determine the input and output data for each sample. For example, take the data at time (n-1) as the input data and the typical behavioral feature data corresponding to the data at time n as the output data.

[0060] Step S303: Process the training data into the input format of the LSTM model. For example, transform a set of user behavior data over a period of time into the input format of the LSTM model: [N, seq_len, features], where N represents the number of samples; seq_len represents the time step size of a sample; and features represents the number of features at each time step, which is also the actual input data of the LSTM at each time point.

[0061] Step S304 defines the constants, fitting function, and structure of the LSTM model. This includes defining the number of input features, the number of neurons in an LSTM unit, the sample size, the time step, the learning rate, the number of training iterations, the output probability, the number of hidden layers (one LSTM unit is placed in each hidden layer), and selecting the optimizer and loss function, etc.

[0062] Step S305 involves training, validating, and evaluating the model, and adjusting the parameters based on the validation results. To ensure effective comparison of the final model, the Adam algorithm is used to dynamically adjust the learning rate of each parameter based on the first and second moment estimates of the gradient of each parameter with respect to the loss function. Since Adam is based on gradient descent, the learning step size of the parameters in each iteration has a defined range, and the parameter values ​​are relatively stable. The evaluation criterion for the model training results is the Mean Absolute Percentage Error (MAPE), whose value is approximately close to 0. Higher prediction accuracy is indicated by formulas 1-7 as follows:

[0063]

[0064] In practical applications, specific parameters need to be adjusted over time and as the amount of training data increases.

[0065] After the above training is completed and the evaluation is satisfactory, the model is saved in step S306. When using the model, inputting a processed time-series dataset of user behavior will yield the prediction result, which is a typical behavioral feature data.

[0066] In one embodiment, the behavioral feature data output by the LSTM neural network model at the last time step is used as the final behavioral feature data.

[0067] In another embodiment, such as Figure 6 The diagram shows the flowchart of a method for obtaining final behavioral feature data based on a behavioral prediction model. It specifically includes the following steps:

[0068] Step S311: When inputting model sample data into the LSTM neural network model, obtain the behavioral feature data output at each time step and use it as the original behavioral feature data.

[0069] Step S312: Cluster the raw behavioral feature data obtained from all time steps.

[0070] Step S313: Determine if the number of behavioral feature data obtained after clustering is 1. If only one behavioral feature data is obtained after clustering, then in step S314, use it as the final behavioral feature data. If multiple intermediate behavioral feature data are obtained after clustering, then in step S315, call the pre-trained DRNN model, use the multiple intermediate behavioral feature data as input to the DRNN (Deep Recurrent Neural Network) model, obtain one behavioral feature data through the DRNN model, and use it as the final behavioral feature data.

[0071] The structure diagram of the DRNN model is as follows: Figures 7A-7B As shown. The DRNN model includes an input layer, a network regression layer, and an output layer, where, therefore, I i (k)I i (k) represents the intermediate behavioral feature data obtained after clustering, which serves as the input to the i-th neuron in the input layer. The output of the j-th neuron in the network regression layer is X. j (k), S j (k) represents the sum of inputs to the j-th regression neuron, f(·) is the S-function, and O(k) is the output of the DRNN network. Specific formulas are shown in Formulas 2-1, 2-3, and 2-3 below.

[0072]

[0073] X j (k)=f(S j (k)) 2-2

[0074]

[0075] In the formula, W I W D and W O The weight vectors of the network input layer, network regression layer, and output layer are respectively represented by W. I For example, it can be represented as:

[0076]

[0077] When training a DRNN model, it is also necessary to determine the training and test sets, where the data in the training and test sets consists of behavioral feature data. Then, the parameters, functions, and DRNN model network structure are defined, and the DRNN model is trained, validated, and evaluated. These details will not be elaborated further here. It is important to note that since human behavior represents a dynamic array of data that changes in real time, it is necessary to utilize both forward and backward propagation of the DRNN neural network for comprehensive learning. Backpropagation calculates the learning bias in reverse along the connection paths, using gradient descent to challenge the weights and thresholds of neurons in each layer, thereby continuously reducing the learning bias.

[0078] Learning through positive input from a multidimensional behavior array I (each multidimensional behavior array represents a behavior feature data point), the input to the network regression layer is Equation 2-3, i.e.:

[0079]

[0080] The output of the network regression layer is given by formula 2-2, i.e., X. j (k)=f(S j (k)), where f(·) takes the double S function, i.e., formula 2-5

[0081]

[0082] The output of the network output layer is: The error between the actual output and the network output is given by formula 2-6:

[0083] e(k)=y(k)-y m (k) 2-6

[0084] In the formula, y(k) is the actual output, y m (k) represents the network output.

[0085] In this embodiment, the performance indicators are taken from formulas 2-7:

[0086]

[0087] When the behavior array I changes over time, a bias is introduced in the learned behavior. Therefore, backpropagation of gradient descent driven by a neural network is used to adjust the weights between layers. The weight learning algorithm is as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094] in:

[0095]

[0096] η in the formula I η D η O, where represents the learning rate of the input layer, regression layer, and output layer, respectively, and is the momentum factor.

[0097] The object's Jacobian information is:

[0098]

[0099] When multiple intermediate behavioral feature data I are used as input to the DRNN model in step S315, a behavioral feature data is obtained through the DRNN model.

[0100] In this embodiment, the clustered data is processed using the DRNN multi-neural network algorithm, which further improves the accuracy of prediction and reduces errors.

[0101] Since the LSTM model can obtain predicted data for each time step (i.e., the aforementioned raw behavioral feature data), and its input data is a collection of multiple time-series data with a certain duration, the above embodiment predicts the user's likely next behavior. In another embodiment, using the raw behavioral feature data obtained from each time step prediction, the specific behavior category of the user within the time period can also be predicted. For example, a classification model is trained based on multiple known behaviors in the metaverse user behavior training system. These user behaviors include various types of typical behaviors, such as running during normal exercise, normal walking, dancing, running while being chased, evading capture, chasing someone, etc. Each typical behavior is a classification category. When predicting the user's behavior category for this time period, multiple raw behavioral feature data obtained from the LSTM model are used as input to the classification model. The classification model calculates the score belonging to the known behavior category according to the classification algorithm, and the behavior category with the highest score is determined as the comprehensive behavior category of the metaverse user in the current time period. For example, when predicting the user's next behavior based on the metaverse user behavior data set for the current time period, the model can predict the user's next behavior, such as... Figure 3F The action shown is running. When the raw behavioral feature data obtained at each time step is used to classify the behavior category of the current time period, it can be determined that the user was running while exercising during the time period, rather than running while being chased.

[0102] The input and output data of the various models in the aforementioned embodiments are multi-dimensional matrices composed of static motion feature data and dynamic body part performance data, specifically described as X = [S, A]. The data processing node in the metaverse space processing system includes a device for acquiring metaverse space user behavior data. This device processes user behavior data within a preset time period to obtain a multi-dimensional matrix composed of static motion feature data and dynamic body part performance data, and encodes it according to body parts, thereby forming a metaverse space personal behavior feature knowledge base.

[0103] Specifically, Figure 8 This is a flowchart of a method for obtaining user behavior data in the metaverse space according to an embodiment of the present invention. Figure 9 This is a schematic diagram of a data structure according to an embodiment of the present invention. (In conjunction with...) Figure 8 and Figure 9 The method for obtaining user behavior data in the metaverse space in this embodiment includes:

[0104] Step S81: Obtain the target user's behavioral performance data within a preset time period from the Metaverse real-time database based on the target user's identity information.

[0105] Step S82: Within the preset time period, determine multiple analysis times T1, T2...T with time sequence. n In one embodiment, the analysis time may be the metaverse space refresh time or a multiple thereof.

[0106] Step S83: Extract static performance data M of various body parts corresponding to the analysis time from the behavioral performance data. X Where X represents a body part, such as M. h Represents the head. For analysis times T1, T2...T n The static head performance data are M h1 M h2 M hn M b M represents the torso. al M represents the left arm. ar M represents the right arm. ll and M lr These represent the left and right legs, respectively. Therefore, the static performance data M for each body part... X =[M X1 M X2 M X3 ...M Xi ...M Xn ], where i represents the analysis time sequence number and n is the number of analysis times.

[0107] Step S84: Combine the static performance data of multiple body parts at each analysis moment to obtain the static motion feature data S at each analysis moment. Ti The data for the head, torso, left arm, right arm, left leg, and right leg at each of the aforementioned analysis moments are combined to form the static motion feature data S for each analysis moment. Ti , where i is the sequence number of the analysis time. For example... Figure 10The action diagram shown is an image reconstructed from a 3D spatial structure dataset. Based on the combined data (3D spatial structure data) of the head, torso, left arm, right arm, left leg, and right leg at time T1, the user's action at time T1 can be determined as follows: Figure 10 As shown. Therefore, the static motion characteristic data S at each analysis time point Ti =[M hi M bi M ali ,……M lri [] represents the static motion characteristics at that moment.

[0108] Step S85: Based on the changes in static performance data of each body part within the preset time period at multiple consecutive analysis moments, dynamic performance data M of the body part is obtained. xd That is: M xd =[M xd1 M xd2 ...M xd(n-1) For example, based on the static head performance data M at time T1 and time T2. h1 M h2 Get T (1-2) Head dynamic performance data M over a period of time hd1 =M h2 -M h1 By analogy, the head dynamic performance data Mhd is obtained, that is: head dynamic performance data M hd =[M hd1 M hd2 ...M hd(n-1) ], where M hd(n-1) =M hn -M h(n-1) .

[0109] Since the static body part performance data is a three-dimensional spatial structure dataset, i.e., a set of three-dimensional coordinates of multiple points, the dynamic performance data is calculated by calculating the difference between the coordinates in the three dimensions. The degree of positional change in three-dimensional space can be obtained through the dynamic performance data. For example, for a dynamic performance data (0, 15, 1), assuming the meta-cosmic spatial coordinate system (x, y, z) corresponds to the x-axis and y-axis used to define the Earth's surface plane, and the z-axis corresponds to the height direction, according to the dynamic performance data (0, 15, 1), the spatial position at two moments differs by 0 and a tiny 1 along the x-axis and y-axis, respectively, and by 15 along the height direction. Therefore, it can be determined that the user is almost vertically upward.

[0110] Step S86, combining multiple static motion feature data S Ti and the corresponding dynamic performance data of body parts Mxd Obtain the user's behavior data X. Where the behavior data X = [S, A], S = [S...]. T1 S T2 ...S Tn ] represents the static motion feature data corresponding to n analysis times, A = [M hd M bd ,……M lrd ] represents the dynamic feature data of all body parts corresponding to n analysis times.

[0111] Through the above processing, the embodiments of the present invention use an encoded multidimensional matrix to describe user behavior in the metaverse space. The data structure is concise and clear, and can comprehensively express user behavior.

[0112] Based on multiple static motion feature data S Ti It is possible to determine the spatial location of each body part in the metaverse at the moment of analysis, such as Figure 10 As shown. Based on the dynamic performance data M of body parts. xd It is possible to determine the degree of change of the same body part in two different spatial locations. Based on these two types of data, the behavior and type of the current user within that time period can be obtained. Figure 10 From time T1 to T 12 The static motion feature data and the corresponding dynamic performance data of the body parts during this time period can be used to obtain the user's "standing long jump" behavior.

[0113] Since the user behavior data in this invention is a metaverse spatial structure dataset, at least three-dimensional data needs to be calculated, resulting in a large amount of computation. To reduce the amount of data processing, in one embodiment, when extracting the static performance data of various body parts corresponding to the analysis time in step S83, only the three-dimensional spatial structure dataset of one or more representative body parts that can represent the whole body part is extracted as the static performance data of the body part. For example, the head is a dataset composed of multiple spatial points. After simplification, the static performance data of the head can be represented by the three-dimensional coordinates of only one center point. For parts with length in a certain dimension, such as the torso and limbs, they can be represented by the three-dimensional coordinates of multiple points, thereby simplifying the computation. Correspondingly, the dynamic performance data of each body part is a set of static difference data obtained from the three-dimensional spatial structure dataset of one part, or multiple sets of static difference data obtained from the three-dimensional spatial structure datasets of multiple body parts respectively.

[0114] Furthermore, when obtaining the static motion feature data for each analysis time in step S84, the difference between the static motion feature data of two adjacent analysis times is compared. If the difference meets the data merging requirements, such as being less than a threshold, the static motion feature data of the two analysis times are merged, and the merged result data is used as the static motion feature data of one analysis time, thereby further reducing the amount of data.

[0115] Furthermore, when obtaining the dynamic performance data of the body parts in step S85, data simplification can also be performed. For example... Figure 11 As shown, Figure 11 This is a flowchart of a method for analyzing the changes in static performance data of a body part over multiple consecutive analysis times according to an embodiment of the present invention to obtain dynamic performance data of the body part, specifically including the following steps:

[0116] Step S851: Calculate the difference in static performance data of the same body part at two adjacent analysis times to obtain the static difference data of the body part. For example, calculate the static performance data M of the head at times T1 and T2. h1 M h2 The difference yields T. (1-2) Time-segment head dynamic performance data M hd1 =M h2 -M h1 Calculate the static head performance data M at time T2 and time T3. h2 M h3 The difference yields T. (2-3) Head dynamic performance data M over a period of time hd2 =M h3 -M h2 .

[0117] Step S852: Compare two adjacent static difference data. For example, calculate M. hd1 With M hd2 The difference ΔM hd1 As a comparison result.

[0118] Step S853: Determine whether the difference between two adjacent static difference data meets the data merging requirements. For example, if a merging threshold is set, then compare M... hd1 With M hd2 The difference ΔM hd1 The value of the set threshold. If ΔM hd1 If the value is less than the threshold, it is considered to meet the data merging requirements. Therefore, in step S854, the two static difference data are merged, and the merged result is treated as a single static difference data. If ΔM hd1 If the data is greater than or equal to the threshold, it is considered not to meet the data merging requirements and will not be merged.

[0119] Step S855: Determine whether all static difference data has been processed. If not, return to step S852. If processed, combine the multiple static difference data to form the dynamic performance data of the body part in step S856.

[0120] In step S853, the threshold set is a three-dimensional threshold, meaning that two static difference data sets meet the data merging requirements when they satisfy the thresholds for each of the three dimensions. The three-dimensional thresholds can be the same or different depending on the user behavior scenario and are set according to the accuracy requirements for each dimension. Merging the two static difference data sets includes discarding one data set and retaining the other, or calculating their average and using the average as the merged static difference data set.

[0121] The aforementioned data simplification process in steps S83-85 saves data storage space and reduces the amount of data processing required during subsequent use.

[0122] In another embodiment of the invention, the dynamic performance data of the body parts are inertial index data characterizing changes in the spatial position of the body parts, such as velocity and / or acceleration in one or more dimensions of three-dimensional space. Figure 11 Taking the dynamic performance data obtained after processing as an example, dividing the static difference data in the dynamic performance data by the time difference between the two moments of the static performance data yields the velocity, an inertial index. Alternatively, dividing the difference between the two velocities by the time difference yields the acceleration, an inertial index. Furthermore, the dynamic performance data can be expressed by one or more functions based on the changing patterns within it. Taking the head static performance data M as an example... h1 M h2 ...M hn For example, since the static performance data of each head is a three-dimensional spatial structure data, the velocity / acceleration or velocity / acceleration function in three dimensions can be obtained based on time and multiple coordinates in each dimension. When the velocity / acceleration in a certain dimension is 0, the dynamic performance data M can be simplified. hdi Velocity / acceleration in a single dimension. For example... Figure 10 In the T1 and T2 data points shown, the head, arms, torso, and legs all exhibit certain velocities / accelerations in all three dimensions. Taking the x-axis as an example, the velocity / acceleration of the arms is significantly greater than that of the head, torso, and legs. Furthermore, the velocity / acceleration of the legs can be broadly categorized into two types: the velocity / acceleration of the feet is zero, and the velocity / acceleration increases the closer the legs are to the torso. In this embodiment, the dynamic performance data is converted into more meaningful indicators to facilitate subsequent data use.

[0123] On the other hand, the present invention also provides a device for predicting user behavior in a metaverse space, such as... Figure 12 The diagram shows a block diagram of a device for predicting user behavior in the metaverse space. It includes a data acquisition module 100, a sample processing module 120, a prediction module 140, and a behavior recognition module 160. The data acquisition module 100 acquires a set of behavioral data for a user in the metaverse space during a preset time period. This behavioral data set is a time-series data set including multiple time-based behavioral datasets. Each time-based behavioral dataset is a multi-dimensional matrix composed of static action feature data and dynamic body part performance data, which serves as the user's behavioral data X, where X = [S, A], S = [S...]. T1 S T2 ...S Tn ] represents the static motion feature data corresponding to n analysis times, where the static motion feature S at the i-th analysis time is... Ti =[M hi M bi M ali ,……M lri ], M hi M represents the static head behavior data at the i-th analysis time. bi For the static trunk performance data at the i-th analysis time, M ali M represents the static performance data of the left arm at the i-th analysis time. lri Let A represent the static performance data of the right leg at the i-th analysis time. hd M bd ,……M lrd ] represents the dynamic feature data of all body parts corresponding to n analysis times. M hd Let M be the head dynamic performance data at the nth analysis time, where M hd =[M hd1 M hd2 ...M hd(n-1) ], M hd(n-1) =M hn -M h(n-1) The sample processing module 120 is connected to the data acquisition module 100 and is used to process the time-series data set into model sample data for multiple time steps, the number of time steps corresponding to the number of moments in the time-series data set. The prediction module 140 is connected to the sample processing module 120 and is used to retrieve the pre-trained behavior prediction model and input the corresponding model sample data into the behavior prediction model according to the time steps to obtain a final behavior feature data. The behavior recognition module 160 is connected to the prediction module 140 and is used to obtain the typical behavior corresponding to the final behavior feature data as the prediction result.

[0124] In one embodiment, such as Figure 13 As shown, the prediction module 140 includes a model unit 141, a data input unit 142, and an output data acquisition unit 143. The model unit 141 is used to retrieve a pre-trained behavior prediction model. The data input unit 142 is connected to the model unit 141 and is used to input corresponding model sample data to the behavior prediction model according to time steps. The output data acquisition unit 143 is connected to the model unit 142 and is used to acquire the behavior feature data output by the behavior prediction model.

[0125] In one embodiment, when the behavior prediction model is an LSTM neural network model, the output data acquisition unit 143 acquires the behavior feature data output by the LSTM neural network model at the last time step and uses it as the final behavior feature data.

[0126] In another embodiment, the prediction module 140 further includes a clustering unit 144 and a DRNN model unit 145, such as Figure 13 As shown by the dashed line, it is connected to the output data acquisition unit 143. The output data acquisition unit 143 acquires the output data of the LSTM neural network model at each time step and sends it to the clustering unit 144. The clustering unit 144 clusters the raw behavioral feature data output by the LSTM neural network model at each time step. When a final behavioral feature data is obtained through clustering, the output data acquisition unit 143 acquires the final behavioral feature data obtained through clustering. When multiple intermediate behavioral feature data are obtained through clustering, a notification is sent to the DRNN model unit 145. The DRNN model unit 145 calls the pre-trained DRNN model, using the multiple intermediate behavioral datasets obtained by the clustering unit 144 as input to the DRNN model. The output data acquisition unit acquires a behavioral dataset output by the DRNN model as the final behavioral feature data.

[0127] Figure 14This is a block diagram illustrating the principle of a behavior recognition module according to an embodiment of the present invention. In one embodiment, the behavior recognition module 160 includes a single behavior confirmation unit 161 and a comprehensive behavior analysis unit 162. The single behavior confirmation unit 161 is connected to the output data acquisition unit 143 in the prediction module 140, and determines a user's single behavior based on typical behaviors corresponding to the final behavior feature data. The comprehensive behavior analysis unit 162 is connected to the output data acquisition unit 143 in the prediction module 140. The output data acquisition unit 143 acquires the raw behavior feature data output by the LSTM neural network model at each time step and sends it to the comprehensive behavior analysis unit 162. The comprehensive behavior analysis unit 162 analyzes the raw behavior feature data and classifies the comprehensive behavior of the user in the metaverse space during the current preset time period to obtain the comprehensive behavior category for the current preset time period.

[0128] When the database of the metaverse space system already contains a user behavior feature library, the data acquisition module 100 reads the behavior data set for the corresponding time period from the user behavior feature library. In another embodiment, when the database of the metaverse space system only contains user behavior performance data, the data acquisition module 100 generates a user behavior data set based on the user behavior performance data.

[0129] Figure 15 This is a schematic diagram of a data acquisition module according to an embodiment of the present invention. The data acquisition module in this embodiment is used to obtain user behavior data in the metaverse space. The data acquisition module 100 includes a data acquisition unit 101, a time segmentation unit 102, a static data extraction unit 103, a static action feature unit 104, a dynamic performance data unit 105, and a behavior data determination unit 106. The data acquisition unit 101 acquires the target user's behavior performance data within a preset time period from the metaverse real-time database based on the target user's identity information. The behavior performance data is a three-dimensional spatial structure dataset representing user behavior, which is a time-series data set.

[0130] The time segmentation unit 102 is connected to the data acquisition unit 101. After the data acquisition unit 101 obtains the behavioral performance data within a preset time period, the time segmentation unit 102 determines multiple analysis times with time sequence within the preset time period. The analysis times can be the metaverse space refresh time or a multiple thereof, and then notifies the static data extraction unit 103.

[0131] The static data extraction unit 103 is connected to the time segmentation unit 102. After receiving a notification from the time segmentation unit 102, it extracts various static performance data of different body parts from the behavioral performance data according to the determined analysis time. For example, each body part static performance data M X =[M X1 M X2 M X3 ...M Xn [where X represents a body part and i represents the analysis time number]. After extracting the data, the static motion feature unit 104 and the dynamic performance data unit 105 are notified.

[0132] The static motion feature unit 104 is connected to the static data extraction unit 103. After receiving a notification from the static data extraction unit 103, it combines static performance data of multiple body parts at each analysis moment to obtain static motion feature data for each analysis moment. For example, the static motion feature data S at each analysis moment... Ti =[M hi M bi M ali ,……M lri ,……], where i is the analysis time, M hi M bi M ali ,……M lri These are the static performance data at the time of analysis. After processing, the static action feature unit 104 notifies the behavior data determination unit 106.

[0133] The dynamic performance data unit 105 is connected to the static data extraction unit 103. After receiving a notification from the static data extraction unit 103, it analyzes the changes in static performance data of each body part at multiple consecutive analysis times within the preset time period to obtain the dynamic performance data of the body part. In one embodiment, the dynamic performance data M of the body part... xd =[M xd1 M xd2 ...M xdi ...M xd(n-1) ]. Where M xd For dynamic performance data of a body part, M xdi For T-based (i-1) Time and T i Static performance data M of the body parts at any given time X(i-1) M Xi Get T (1-2)Dynamic performance data for a given time period. After processing, the dynamic performance data unit 105 notifies the behavior data determination unit 106.

[0134] The behavior data determination unit 106 is connected to the static motion feature unit 104 and the dynamic performance data unit 105. Upon receiving notifications from the static motion feature unit 104 and the dynamic performance data unit 105, it combines multiple static motion feature data and corresponding body part dynamic performance data to obtain the user's behavior data. For example, the obtained behavior data X = [S, A], S = [S...]. T1， S T2 ...S Tn ] represents the static motion feature data corresponding to n analysis times, A = [M hd M bd ,……M lrd ] represents the dynamic feature data of all body parts corresponding to n analysis times.

[0135] Figure 16 This is a block diagram illustrating the principle of a data acquisition unit according to an embodiment of the present invention. The data acquisition unit 101 further includes an identity information unit 1011, a data reading unit 1012, and a data refinement unit 1013. The identity information unit 1011 acquires the identity information of a target user and sends this information to the data reading unit 1012. The data reading unit 1012, connected to the identity information unit 1011, acquires a dataset of the three-dimensional spatial structure of various body parts of the target user within a preset time period from a real-time metaverse database based on the target user's identity information, and notifies the data refinement unit 1013. The data refinement unit 1013, connected to the data reading unit 1012, upon receiving the acquired data, retrieves a dataset of one or more three-dimensional spatial structures representing the entire body part from the dataset of the three-dimensional spatial structure of each body part as static representation data of the body part. For example, the three-dimensional coordinates representing the center point of the head can be used as static head representation data, and the three-dimensional coordinates representing multiple points of the torso, limbs, etc. can be used as static head representation data, thereby simplifying the calculations in subsequent steps.

[0136] Figure 17This is a block diagram illustrating the principle of a static motion feature unit according to an embodiment of the present invention. In this embodiment, the static motion feature unit 104 includes a first data combination unit 1041, a motion comparison unit 1042, and a first merging unit 1043. The first data combination unit 1041 combines static performance data of multiple body parts at each analysis time to obtain static motion feature data for each analysis time, and then sends a notification to the motion comparison unit 1042. The motion comparison unit 1042 is connected to the data combination unit 1041. After receiving the notification from the data combination unit 1041, it compares the differences in static motion feature data between two adjacent analysis times; when the differences meet the data merging requirements, it sends a merging notification to the first merging unit 1043. The first merging unit 1043 is connected to the motion comparison unit 1042. Upon receiving the merging notification from the motion comparison unit 1042, it merges the static motion feature data from the two analysis times, and uses the merged result data as static motion feature data for one analysis time. In this embodiment, the static motion feature data at two feature moments are not significantly different, which further reduces the amount of data.

[0137] Figure 18 This is a block diagram illustrating the principle of a dynamic performance data unit according to an embodiment of the present invention. In this embodiment, the dynamic performance data unit 105 includes a difference calculation unit 1051, a difference comparison unit 1052, a second merging unit 1053, and a second data combination unit 1054. Specifically, when the difference calculation unit 1051 receives a notification from the static data extraction unit 103, it calculates the difference in static performance data of the same body part at two adjacent analysis times to obtain static difference data for the body part, and then notifies the difference comparison unit 1052 after calculation. The difference comparison unit 1052 is connected to the difference calculation unit 1051 and is used to compare two adjacent static difference data. When the difference between two adjacent static difference data meets the data merging requirements, a merging notification is sent to the second merging unit 1053. The second merging unit 1053 is connected to the difference comparison unit 1051, and after receiving the merging notification, it merges the two static difference data, using the merged result as a single static difference data. The second data combination unit 1054 is connected to the second merging unit 1053 to combine multiple static difference data to form the dynamic performance data of the body part. In this embodiment, the amount of data is further reduced by merging the static difference data.

[0138] After processing by the data acquisition module 100, a multidimensional matrix with encoding is obtained to describe user behavior in the metaverse space. The data structure is concise and clear, and can comprehensively express user behavior. After the data acquisition module 100 obtains the multidimensional matrix composed of static action feature data and corresponding dynamic performance data of body parts for a preset time period, it sends it to the sample processing module 120 to obtain samples for model prediction. After processing by the prediction module 140, a multidimensional matrix with typical behaviors is obtained, and the corresponding behaviors can be obtained through the behavior recognition module.

[0139] The present invention also provides an electronic device, comprising: a processor and a memory storing computer program instructions; wherein the processor executes the computer program instructions to implement the prediction method for predicting user behavior according to any of the above embodiments.

[0140] Figure 19 A schematic diagram of the hardware structure of an embodiment of the electronic device provided by the present invention is shown.

[0141] like Figure 19 As shown, the electronic device may include a processor 601 and a memory 602 storing computer program instructions.

[0142] Specifically, the processor 601 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement embodiments of the present invention.

[0143] Memory 602 may include mass storage for data or instructions. For example, and not limitingly, memory 602 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 602 may include removable or non-removable (or fixed) media. Where appropriate, memory 602 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 602 is non-volatile solid-state memory.

[0144] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform predictive operations of user behavior as described with reference to one aspect of the invention.

[0145] The processor 601 reads and executes computer program instructions stored in the memory 602 to implement any of the user behavior prediction methods in the above embodiments.

[0146] In one example, the electronic device may also include a communication interface 603 and a bus 610. For example, Figure 19 As shown, the processor 601, memory 602, and communication interface 603 are connected via bus 610 and communicate with each other. The electronic device in this embodiment of the invention can be a server or other computing device, or it can be a cloud server.

[0147] The communication interface 603 is mainly used to realize communication between various units, devices, units and / or equipment in the embodiments of the present invention.

[0148] Bus 610 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 610 may include one or more buses. While specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0149] Furthermore, in conjunction with the product search method in the above embodiments, this invention can be implemented using a computer storage medium. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the user behavior prediction methods described in the above embodiments.

[0150] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0151] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Machine-readable media can include non-transitory computer-readable storage media, such as electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, and can also include radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0152] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0153] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0154] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A method for predicting user behavior in a metaverse space, comprising: Obtain a set of behavioral data of a user in the metaverse space for the current preset time period, wherein the set of behavioral data is a time-series set including behavioral data at multiple moments, and the behavioral data at each moment includes static action feature data and dynamic performance data of body parts; The time series data set is processed into model sample data for multiple time steps; The pre-trained behavior prediction model is retrieved, and corresponding model sample data is input into the behavior prediction model according to time steps to obtain a final behavior feature data. The behavior prediction model is an LSTM neural network model. The typical behavior corresponding to the final behavioral feature data is obtained as the prediction result; The behavioral feature data output by the LSTM neural network model at each time step is obtained as the original behavioral feature data; multiple original behavioral feature data are clustered, and a behavioral feature data is obtained after clustering, which is used as the final behavioral feature data; and in response to multiple intermediate behavioral feature data obtained after clustering, the pre-trained DRNN model is called, with the multiple intermediate behavioral feature data as the input of the DRNN model, and a behavioral feature data output by the DRNN model as the final behavioral feature data.

2. The method of claim 1, further comprising: Analyze multiple raw behavioral feature data to determine the comprehensive behavioral category of metaverse space users during the current preset time period.

3. The method according to claim 2, wherein, When analyzing multiple raw behavioral feature data, a pre-trained classification model is invoked, using the multiple raw behavioral feature data as input to the classification model. The behavioral category with the highest score from the classification model is determined as the comprehensive behavioral category of the user in the metaverse space during the current preset time period.

4. The method according to claim 1, wherein the training dataset of the behavior prediction model is a set of behavior data of known behaviors of users in the metaverse space during a preset time period, read from the metaverse real-time database; the sample data is moment behavior data composed of static action feature data and dynamic performance data of body parts; the label of the sample is the known behavior; and the output of the model is the behavior feature data corresponding to the known behavior.

5. The method according to claim 1, wherein the step of obtaining the behavioral data set of the user in the metaverse space during the current preset time period includes: Based on the user's identity information in the Metaverse space, the system reads the target user's behavioral data set for the current preset time period from the Metaverse real-time database.

6. The method according to claim 1, wherein the step of obtaining the behavioral data set of the user in the metaverse space during the current preset time period includes: Acquire behavioral data of users in the metaverse space during the current preset time period; Within the preset time period, multiple analysis moments are determined in chronological order; Extract static performance data of various body parts corresponding to the analysis time from the behavioral performance data; By combining the static performance data of multiple body parts at each analysis moment, the static motion characteristic data of each analysis moment are obtained. Analyze the changes in static performance data of each body part at multiple consecutive analysis times within the preset time period to obtain dynamic performance data of the body part; and By combining multiple static motion feature data and corresponding dynamic performance data of body parts, a set of user behavior data is obtained.

7. A device for predicting user behavior in a metaverse space, comprising: The data acquisition module is configured to acquire a set of behavioral data of the user in the metaverse space during the current preset time period. The behavioral data set is a time-series data set that includes behavioral data at multiple moments. The behavioral data at each moment includes static action feature data and dynamic performance data of body parts. A sample processing module, which is connected to the data acquisition module, is configured to process the time series data set into model sample data for multiple time steps; A prediction module, connected to the sample processing module, is configured to retrieve a pre-trained behavior prediction model and input corresponding model sample data into the behavior prediction model according to time steps to obtain a final behavior feature data. The behavior prediction model is an LSTM neural network model. A behavior recognition module, which is connected to the prediction module, is configured to obtain the typical behavior corresponding to the final behavior feature data as the prediction result; The prediction module includes: The model unit is configured to retrieve a pre-trained behavior prediction model. A data input unit, which is connected to the model unit, is configured to input corresponding model sample data into the behavior prediction model according to time steps; An output data acquisition unit, which is connected to the model unit, is configured to acquire behavioral feature data output by the behavior prediction model; A clustering unit, connected to the model unit, is configured to cluster the raw behavioral feature data output by the LSTM neural network model at each time step, obtaining one final behavioral feature data or multiple intermediate behavioral feature data; wherein, in response to obtaining a behavioral feature data after clustering, it is used as the final behavioral feature data; and The DRNN model unit, which is connected to the clustering unit and the output data acquisition unit, calls the pre-trained DRNN model in response to obtaining multiple intermediate behavioral feature data after clustering and classification. The multiple intermediate behavioral data obtained by the clustering unit are used as the input of the DRNN model, and the output data acquisition unit obtains one behavioral data output by the DRNN model as the final behavioral feature data.

8. The apparatus according to claim 7, wherein, When the behavior prediction model is an LSTM neural network model, the output data acquisition unit acquires the behavior feature data output by the LSTM neural network model at the last time step and uses it as the final behavior feature data.

9. The apparatus of claim 7, wherein the behavior recognition module further comprises: A single behavior confirmation unit is configured to determine a corresponding single typical behavior based on behavior feature data, wherein the behavior feature data includes final behavior feature data. and The comprehensive behavior analysis unit, which is connected to the model unit, is configured to analyze the raw behavioral feature data output by the LSTM neural network model to obtain the comprehensive behavior category of the metaverse space user in the current preset time period.

10. An electronic device comprising: Processor and memory storing computer program instructions; When the electronic device executes the computer program instructions, it implements the method as described in any one of claims 1-6.

11. A computer-readable storage medium, wherein, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the method as described in any one of claims 1-6.

12. A computer program product comprising computer program instructions that, when executed by a processor, implement the method as described in any one of claims 1-6.

Images