A Metaverse Interactive Scene Management System Based on Deep Learning

Abstract

The present invention relates to a metaverse interaction scene management system based on deep learning, which collects user behavior characteristics and corresponding metaverse interaction scene data samples in daily work; sorts the importance of each element of the metaverse interaction scene by constructing a random forest model between different user behavior characteristics and metaverse interaction scenes; and deletes the metaverse interaction scene elements with an importance order lower than n according to the importance order of each element of the metaverse interaction scene. The present invention solves the problem that the prior art is difficult to manage according to the importance of different metaverse interaction scene elements, resulting in a large amount of redundant data occupying memory, which reduces the operating efficiency of the metaverse.

Description

Technical Field

[0001] The present invention belongs to the field of metaverse technology and relates to a metaverse interactive scene management system based on deep learning. Background Art

[0002] The metaverse is a virtual, interactive world that exists in digital form and contains a large number of users and entities. The metaverse can integrate the real world with the virtual world and produce a new form of Internet. In the metaverse, people can use virtual reality technology to interact with others, participate in various activities and games, shop, education, medical care, etc.

[0003] At present, the Metaverse technology can match the corresponding interaction scenarios according to the user's behavior characteristics. However, due to the different behavior characteristics and preferences of users, the different Metaverse interaction scenarios matched contain a large number of scene elements, which creates huge storage pressure. The existing technology is difficult to manage the importance of different Metaverse interaction scene elements, resulting in a large amount of redundant data occupying the memory, which greatly reduces the efficiency of the Metaverse operation. Summary of the invention

[0004] The present invention provides a metaverse interaction scene management system based on deep learning, which aims to collect large-scale user behavior characteristics and corresponding matching metaverse interaction scene data samples in daily work, establish a random forest model between user behavior characteristics and metaverse interaction scenes, sort the importance of each element of the metaverse interaction scene, and delete the scene elements with lower importance, so as to solve the problem that the existing technology is difficult to manage according to the importance of different metaverse interaction scene elements, resulting in data redundancy and reduced metaverse operation efficiency.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present application provides a metaverse interactive scene management system based on deep learning, including a collection module, an analysis module and a management module, wherein the collection module, the analysis module and the management module are communicatively connected, wherein:

[0007] The acquisition module is used to collect user behavior characteristics and their corresponding Metaverse interaction scene data samples in daily work;

[0008] The analysis module is used to rank the importance of each element of the Metaverse interaction scenario by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario;

[0009] The management module is used to delete the metaverse interaction scene elements whose importance order is lower than n according to the importance order of each element of the metaverse interaction scene.

[0010] Furthermore, the user behavior characteristics include body movements, voice communication, sound characteristics, facial expressions and eye movement trajectories.

[0011] Furthermore, the elements of the metaverse interaction scene include objects, roles, functions, environments and events.

[0012] Furthermore, the importance of each element of the Metaverse interaction scenario is ranked by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario, including the following steps:

[0013] S1. Determine the user behavior category based on the similarity between different user behavior characteristics;

[0014] S2. Obtaining the element matrix of the Metaverse interaction scene through the Metaverse interaction scene data;

[0015] S3. According to the element matrix, with the metaverse interaction scene elements as features and the user behavior categories as labels, a random forest model is constructed to rank the importance of the metaverse interaction scene elements.

[0016] Furthermore, the element matrix has user behavior categories as rows and metaverse interaction scene elements as columns.

[0017] Furthermore, determining the user behavior category according to the similarity between different user behavior features includes the following steps:

[0018] S11, obtaining image features of user behavior features, wherein the image features include pixel values, textures, and shapes;

[0019] S12. Using a K-means clustering method, the data samples are divided into a number of user behavior categories according to the image features.

[0020] Furthermore, the K-means clustering method comprises the following steps:

[0021] Normalize the pixel values ​​of image features;

[0022] Determine the appropriate number of cluster centers Q, and select Q data objects as initial cluster centers;

[0023] The data objects with the smallest Euclidean distance from the initial cluster center are taken as a set to form an initial cluster;

[0024] Calculate the average value of the data objects of each initial cluster as the new cluster center;

[0025] The cluster center is updated repeatedly until the cluster center no longer changes, and the final cluster is determined; data objects with the same final cluster are divided into the same user behavior category.

[0026] Furthermore, in step S3, the construction of the random forest model to rank the importance of the elements of the metaverse interaction scene includes the following steps:

[0027] S31, randomly select 70% of the data objects from the data set for modeling each time;

[0028] S32, randomly extracting a number of metaverse interaction scene elements as explanatory variables and user behavior categories as response variables, and generating a decision tree;

[0029] S33, repeatedly generate a number of decision trees that traverse all metaverse interaction scene elements to generate a random forest;

[0030] S34, calculating the prediction error of each decision tree in the random forest for the data objects not used for modeling;

[0031] S35. Sort the importance of the variables of the metaverse interaction scene elements according to the prediction error of the random forest for the out-of-bag data.

[0032] Furthermore, in step S33, the number of decision trees is determined by the following steps:

[0033] With a step size of 5 decision trees, 20 random forests with 5 to 100 decision trees were generated;

[0034] Calculate the prediction errors of 20 random forests for data objects that are not used for modeling;

[0035] Select the random forest with the smallest prediction error and determine the number of decision trees.

[0036] Furthermore, according to the order of importance of each element of the metaverse interaction scene, the metaverse interaction scene elements with an order of importance lower than n are deleted, and n is the first 50% of the metaverse interaction scene elements in the order of importance.

[0037] Beneficial effects of the present invention:

[0038] By collecting user behavior characteristics and their corresponding Metaverse interaction scene data samples in daily work; by constructing a random forest model between different user behavior characteristics and Metaverse interaction scenes, the importance of each element of the Metaverse interaction scene is ranked; according to the order of importance of each element of the Metaverse interaction scene, the Metaverse interaction scene elements with an order of importance lower than n are deleted. The present invention solves the problem that the prior art is difficult to manage according to the importance of different Metaverse interaction scene elements, resulting in a large amount of redundant data occupying the memory, which reduces the efficiency of the Metaverse operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0040] Figure 1 This is a structural diagram of a metaverse interactive scene management system based on deep learning in the present invention.

[0041] Figure 2 This is a flowchart for sorting the importance of each element of a metaverse interaction scene in one embodiment of the present invention.

[0042] Figure 3 A flowchart of constructing a random forest model in one embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0044] See also Figure 1-Figure 3 The present application provides a metaverse interactive scene management system based on deep learning, including a collection module, an analysis module and a management module, wherein the collection module, the analysis module and the management module are communicatively connected, wherein:

[0045] The acquisition module is used to collect user behavior characteristics and their corresponding Metaverse interaction scene data samples in daily work;

[0046] In this embodiment, the acquisition module aims to systematically collect the behavioral characteristics of users in the metaverse and their corresponding interactive scene data. The module records the performance of users in specific scenarios by real-time monitoring of their behavior, and captures the characteristic information of the scenarios, such as scenario types, element settings, and user configurations. The data collection method includes client logging and user feedback mechanisms to ensure efficient and realistic user experience. The data is first cached locally to ensure integrity when the network is unstable, and then uploaded to the cloud database regularly, using a reliable transmission protocol to ensure security. The module also performs data preprocessing, cleans up missing values ​​and abnormal data, and unifies the format for subsequent analysis. In addition, the monitoring and evaluation function provides a visual interface to track user behavior data and its trends in real time, and regularly evaluates data quality to ensure accuracy. In order to protect user privacy, relevant laws and regulations will be followed during the collection process, and the data will be anonymized. Ultimately, the data accumulated by the acquisition module will provide basic support for the construction of user portraits and the development of intelligent recommendation systems, and promote personalized metaverse experience. In this way, through comprehensive and effective data collection, the module not only improves the user experience, but also lays a solid foundation for subsequent data analysis and application.

[0047] Furthermore, the user behavior characteristics include body movements, voice communication, sound characteristics, facial expressions and eye movement trajectories.

[0048] In this embodiment, user behavior characteristics may include the following aspects:

[0049] Body movements: The user’s body movements and gestures in the Metaverse, such as waving, nodding, squatting, or walking. These movements can reflect the user’s emotions and level of engagement.

[0050] Voice communication: The voice content used by users during the interaction, including speaking speed, tone, emotional expression, etc. This information helps analyze the user's emotional state and social behavior.

[0051] Sound characteristics: The characteristics of the sound produced by the user, such as volume, sound quality, and background noise. These characteristics can affect the user's communication effect and immersive experience.

[0052] Facial expressions: Changes in the user’s facial expressions in the virtual environment, such as smiling, frowning, or surprise, can help determine the user’s emotional response.

[0053] Eye tracking: where the user's eyes move and what they look at, can reveal the user's attention and interest in specific elements.

[0054] It should be noted that in order to match the corresponding Metaverse interaction scene according to the user's body, voice and sound features, the following steps can be performed. First, the user's behavior data is collected in real time through sensors, microphones and cameras to obtain their body movements, voice content and sound features. Next, the data is analyzed to identify body movements (such as waving, pointing), voice emotions (such as happiness or anger) and sound environment (such as noisy or not). Then, the machine learning model is used to match the user's characteristics with the existing interaction scene to identify the most suitable activity type, such as social games or virtual meetings. During the matching process, the model learns the user's preferences and behavior patterns through historical data. In order to improve the user experience, the system also needs to monitor the user's behavior changes in the scene in real time, and dynamically adjust the interaction scene according to their feedback to ensure that they always feel comfortable and engaged. At the same time, personalized recommendations can be provided to users based on their historical behavior and interests to improve participation and satisfaction. This comprehensive use of user behavior characteristics will effectively improve the adaptability of the interaction scene in the Metaverse, make the user experience more immersive and rich, and thus promote long-term user participation.

[0055] Furthermore, the elements of the metaverse interaction scene include objects, roles, functions, environments and events.

[0056] In this embodiment, the metaverse interactive scene elements refer to the various components that constitute the metaverse scene, which can be objects, roles, functions, events, etc., used to present and interact with the virtual environment of the metaverse. Scene elements include but are not limited to:

[0057] 1) Objects: refer to visible and tangible entities in the metaverse, such as buildings, furniture, props, decorations, etc. They add specific visual and tactile experiences to the scene;

[0058] 2) Role: A virtual entity that represents a user or other virtual participant. Roles can be created and managed by users themselves, or they can be virtual characters preset by the system. They play specific roles and functions in the metaverse;

[0059] 3) Function: refers to the various services and functions provided in the Metaverse scene, such as social platforms, game tasks, mall shopping, online learning, work collaboration, etc. Functional elements provide users with tools, services and activities for interaction and use in various scenes;

[0060] 4) Environment: describes the background, topography, climate and other characteristics of the metaverse scene, which create a specific environmental atmosphere for the scene, such as cities, forests, mountains, interstellar space, etc.

[0061] 5) Events: These are specific dynamic situations that occur in the Metaverse scene, such as competitions, performances, gatherings, tasks, plots, etc. These events add rich interactivity and storytelling to the scene.

[0062] The combination and design of scene elements directly affect the richness of the metaverse scene and user experience. Through reasonable arrangement of elements and interaction methods, an attractive and diverse virtual environment can be created.

[0063] The analysis module is used to rank the importance of each element of the Metaverse interaction scenario by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario;

[0064] In this embodiment, the analysis module aims to evaluate and sort the importance of each element in the metaverse interaction scene by constructing a random forest model. The elements of these interactive scenes include objects, roles, functions, environments, and events, which together constitute the user's experience in the metaverse. At the same time, user behavior characteristics include body movements, voice communication, sound characteristics, facial expressions, and eye movement trajectories, which can reflect the user's actual interaction methods and emotional responses in the virtual environment. By associating user behavior characteristics with metaverse elements, the random forest model can identify which elements play a more critical role in user interaction. Ultimately, this analysis can not only provide valuable insights for developers and designers to help them understand the real needs of users, but also guide the optimization and improvement of the metaverse, enhance user immersion and overall satisfaction, so as to better attract and retain users.

[0065] The random forest model is an ensemble learning algorithm that is widely used in classification and regression tasks. It improves accuracy and robustness by building multiple decision trees and combining their predictions. Each decision tree uses randomly sampled data and features during training, which reduces the risk of overfitting and enhances the generalization ability of the model. Random forests can also evaluate the importance of each feature to the prediction results, which provides strong support for feature selection and data analysis. In addition, it can effectively handle high-dimensional data and is suitable for missing values ​​and nonlinear relationships, making it widely used in many fields such as finance, medical care, image recognition, VR, etc. Through this data-driven approach, random forests provide a solid foundation for the design and improvement of metaverse environments.

[0066] Furthermore, the importance of each element of the Metaverse interaction scenario is ranked by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario, including the following steps:

[0067] S1. Determine the user behavior category based on the similarity between different user behavior characteristics;

[0068] In this embodiment, determining the user behavior category based on the similarity between user behavior features is an important method for deeply understanding user needs and optimizing experience. First, through data collection, relevant image features are extracted from the user's behavior in a specific environment, and these features constitute a multidimensional representation of user behavior. Next, the similarity between different user behavior features (image features) is calculated through similarity measurement methods (such as Euclidean distance, cosine similarity, etc.), thereby revealing potential behavior patterns. Subsequently, a clustering algorithm (such as K-means, hierarchical clustering, or DBSCAN) is applied to classify these users and group similar user behaviors into one category. The results of cluster analysis can help identify natural groupings of user behavior, and these groups may exhibit similar needs or preferences. After obtaining the user behavior categories, further analysis can be performed to label each category and understand its characteristics.

[0069] Furthermore, determining the user behavior category according to the similarity between different user behavior features includes the following steps:

[0070] S11, obtaining image features of user behavior features, wherein the image features include pixel values, textures, and shapes;

[0071] In this embodiment, obtaining image features of user behavior features generally includes extracting multiple aspects of the image to comprehensively reflect the user's behavior category.

[0072] Pixel feature analysis includes building a color histogram and calculating brightness and contrast. These features reflect the overall distribution and brightness information of colors in an image. Through these data, emotions and states related to user behavior can be identified. For example, specific color combinations and brightness are related to user anxiety or happiness, which helps understand the user's situation.

[0073] Texture features are usually extracted using local binary patterns (LBP) and Gabor filtering, which can effectively capture subtle texture information in the surrounding area of ​​the image. By analyzing texture features, it is possible to identify changes in user behavior in specific environments, such as sensing whether the user is focused or relaxed, thereby providing deeper insights into user behavior analysis.

[0074] Shape features are mainly obtained through edge detection and shape descriptors, which can help identify the contours and shape features of objects in images. Specifically, edge detection algorithms can capture the boundary positions of objects, while shape descriptors such as Hu invariant moments provide the shape features of objects. This information can be used to identify specific user actions, such as grabbing, waving, etc.

[0075] In addition, if the analysis is a dynamic video, the extraction of motion features is crucial, and the optical flow method can be used to capture the motion trajectory of objects in the image sequence. This method can analyze the speed and direction of the user's movements, thereby providing specific information about the user's behavior, for example, determining whether the user is walking, running, or standing still, and then understanding their activity status.

[0076] S12. Using a K-means clustering method, the data samples are divided into a number of user behavior categories according to the image features.

[0077] Furthermore, the K-means clustering method comprises the following steps:

[0078] Normalize the pixel values ​​of image features;

[0079] In this embodiment, standardization is an important step in data preprocessing, especially when applying algorithms such as K-means clustering. The main purpose of standardization is to eliminate the influence of dimensions and scales between different features, so that each feature has a relatively equal weight in the clustering process. For image features, pixel values ​​are usually in different ranges, such as the pixel value range of grayscale images is 0 to 255, and the value of each color channel in a color image may also be different. This difference may cause certain features to dominate the distance calculation, thereby affecting the final clustering results. Commonly used standardization methods include Z-score standardization and Min-Max scaling. Z-score standardization converts data into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation, which is suitable for situations where there are certain assumptions about the data distribution. Min-Max scaling scales the feature values ​​to a specific range, generally between 0 and 1, which is suitable for situations where there is no obvious distribution assumption but the relative relationship of the data needs to be maintained. In the field of image processing, standardization not only improves the performance of clustering algorithms, but also maintains the consistency of algorithms under different conditions and data sets. Therefore, before performing K-means clustering, standardization is an important step to ensure clustering effect and data comparability. Through standardization, image features can participate in the clustering process in a coordinated manner, thereby improving the accuracy and reliability of the analysis.

[0080] Determine the appropriate number of cluster centers Q, and select Q data objects as initial cluster centers;

[0081] In this embodiment, when initializing the cluster centers, a common practice is to randomly select several data objects in the data space as the initial cluster centers. The reason for this is that by randomly selecting the initial cluster centers, it can be ensured that the algorithm will not be affected by the position of a specific initial cluster center, thereby avoiding falling into a local optimal solution. By randomly selecting data objects in the data space as the initial cluster centers, the algorithm can have a certain degree of randomness, which helps to explore different aspects of the data, and can increase the robustness and reliability of the algorithm, and improve the quality of the final clustering results. Randomly selecting data objects as initial cluster centers is a common and effective initialization method.

[0082] The data objects with the smallest Euclidean distance from the initial cluster center are taken as a set to form an initial cluster;

[0083] Calculate the average value of the data objects of each initial cluster as the new cluster center;

[0084] The cluster center is updated repeatedly until the cluster center no longer changes, and the final cluster is determined; data objects with the same final cluster are divided into the same user behavior category.

[0085] In this embodiment, the cluster center is repeatedly updated until the cluster center no longer changes. The process of determining the final cluster cluster is a core step of the clustering algorithm (such as K-means). After the initial cluster cluster is formed, each cluster will have a representative cluster center. Then, by calculating the distance from each data object to the cluster center, these objects are assigned to the nearest cluster center. After that, the center of each cluster needs to be recalculated, and this new center is the average value of the feature values ​​of all objects in the cluster. This process is iterated continuously until the change of the cluster center is less than the preset threshold or the cluster label no longer changes, indicating that the clustering result has stabilized. This means that the distribution of data objects has become consistent and the cluster center is no longer updated. The purpose of determining the final cluster cluster is to obtain a stable and representative organizational structure so that similar data in the feature space are classified into the same category, thereby effectively revealing the potential patterns and characteristics of the data.

[0086] S2. Obtaining the element matrix of the Metaverse interaction scene through the Metaverse interaction scene data;

[0087] Furthermore, the element matrix has user behavior categories as rows and metaverse interaction scene elements as columns.

[0088] S3. According to the element matrix, with the metaverse interaction scene elements as features and the user behavior categories as labels, a random forest model is constructed to rank the importance of the metaverse interaction scene elements.

[0089] Random forests train and predict by building multiple decision trees, each of which uses only a portion of the training data and a portion of the features for modeling. By building multiple decision trees, a powerful classification or regression model is ultimately obtained, in which each decision tree has a weighted contribution. Random forests can give an importance ranking of each input feature in the model. The importance scores of these features are calculated based on their influence on the model prediction results in random forests. Usually, the feature importance scores in random forests are calculated based on the out-of-bag error.

[0090] Furthermore, in step S3, the construction of the random forest model to rank the importance of the elements of the metaverse interaction scene includes the following steps:

[0091] S31, randomly select 70% of the data objects from the data set for modeling each time;

[0092] S32, randomly extracting a number of metaverse interaction scene elements as explanatory variables and user behavior categories as response variables, and generating a decision tree;

[0093] S33, repeatedly generate a number of decision trees that traverse all metaverse interaction scene elements to generate a random forest;

[0094] S34, calculating the prediction error of each decision tree in the random forest for the data objects not used for modeling;

[0095] It should be noted that out-of-bag data (data objects not used for modeling) refers to data that is not used for model training and has never been used in training. In machine learning and statistical modeling, out-of-bag data is one of the important indicators used to evaluate the performance and generalization ability of the model. The role of out-of-bag data is to test the performance of the model when facing unseen samples. During the training process, the model only uses part of the data set for parameter estimation and model fitting, while out-of-bag data provides an independent data set that can be used to evaluate the prediction accuracy and reliability of the model. In some machine learning algorithms, such as random forests, gradient boosting trees, cross-validation, etc., by using out-of-bag data for model evaluation and tuning, the performance of the model can be better estimated and help identify problems such as overfitting or underfitting.

[0096] S35. Sort the importance of the variables of the metaverse interaction scene elements according to the prediction error of the random forest for the out-of-bag data.

[0097] Furthermore, in step S33, the number of decision trees is determined by the following steps:

[0098] With a step size of 5 decision trees, 20 random forests with 5 to 100 decision trees were generated;

[0099] Calculate the prediction errors of 20 random forests for data objects that are not used for modeling;

[0100] Select the random forest with the smallest prediction error and determine the number of decision trees.

[0101] In the process of building random forests, the number of decision trees used has an important impact on the prediction performance of the model. Here, 20 random forests with 5 to 100 decision trees are generated with a step size of 5 decision trees. This means that 20 random forests with different numbers of decision trees will be created, each containing 5, 10, 15, ..., 100 decision trees. After generating these random forests, predictions are made using data objects that have not been used for modeling, and the prediction error of each random forest is calculated. The prediction error can be evaluated by various indicators, such as mean squared error (MSE) or classification error rate, depending on whether the problem type is regression or classification. The core of this process is that by continuously increasing or decreasing the number of decision trees, the performance changes of the model can be observed. After the prediction error calculations of all 20 random forests are completed, the random forest with the smallest prediction error is selected as the final model. The benefit of this is that it can help people find a balance between preventing overfitting and underfitting, ensuring that the model can maintain good generalization ability for new data. In the end, not only the optimal number of decision trees is determined, but also a proven and efficient model is obtained. This approach effectively exploits the random forest’s reliance on randomness and strong predictive power, ensuring that the selected model can provide excellent predictive performance in practical applications.

[0102] The management module is used to delete the metaverse interaction scene elements whose importance order is lower than n according to the importance order of each element of the metaverse interaction scene.

[0103] In this embodiment, to delete elements below a certain importance threshold n according to the importance order of each element of the metaverse interaction scene, the following steps can be followed:

[0104] First, list all relevant interactive elements for a certain user behavior category; then, through the above data analysis and feedback, evaluate the importance of each element and give a corresponding score. Set an importance threshold n, which may be the top 50% of elements or based on specific scoring criteria. Then, filter out elements with an importance score lower than n and delete them. Finally, verify the effectiveness of the remaining elements through user testing to ensure that they still meet user needs. This process helps to optimize the interaction design of the metaverse, making it more focused on improving the user experience, thereby creating a more attractive and interactive virtual environment.

[0105] Beneficial effects of the present invention:

[0106] By collecting user behavior characteristics and their corresponding Metaverse interaction scene data samples in daily work; by constructing a random forest model between different user behavior characteristics and Metaverse interaction scenes, the importance of each element of the Metaverse interaction scene is ranked; according to the order of importance of each element of the Metaverse interaction scene, the Metaverse interaction scene elements with an order of importance lower than n are deleted. The present invention solves the problem that the prior art is difficult to manage according to the importance of different Metaverse interaction scene elements, resulting in a large amount of redundant data occupying the memory, which reduces the efficiency of the Metaverse operation.

[0107] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A metaverse interactive scene management system based on deep learning, characterized by: It includes a collection module, an analysis module and a management module, which are communicatively connected to each other, wherein: The acquisition module is used to collect user behavior characteristics and their corresponding Metaverse interaction scene data samples in daily work; The analysis module is used to rank the importance of each element of the Metaverse interaction scenario by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario; The management module is used to delete the metaverse interaction scene elements whose importance order is lower than n according to the importance order of each element of the metaverse interaction scene.

2. A metaverse interactive scene management system based on deep learning according to claim 1, characterized in that: The user behavior characteristics include body movements, voice communication, sound characteristics, facial expressions and eye movement trajectories.

3. According to a deep learning-based metaverse interactive scene management system according to claim 1, it is characterized by: The elements of the metaverse interaction scene include objects, roles, functions, environments and events.

4. According to a deep learning-based metaverse interactive scene management system according to claim 1, it is characterized by: The method of ranking the importance of each element of the Metaverse interaction scenario by constructing a random forest model between different user behavior characteristics and the Metaverse interaction scenario includes the following steps: S1. Determine the user behavior category based on the similarity between different user behavior characteristics; S2. Obtaining the element matrix of the Metaverse interaction scene through the Metaverse interaction scene data; S3. According to the element matrix, with the metaverse interaction scene elements as features and the user behavior categories as labels, a random forest model is constructed to rank the importance of the metaverse interaction scene elements.

5. A metaverse interactive scene management system based on deep learning according to claim 4, characterized in that: The element matrix has user behavior categories as rows and metaverse interaction scene elements as columns.

6. A metaverse interactive scene management system based on deep learning according to claim 4, characterized in that: Determining the user behavior category according to the similarity between different user behavior features includes the following steps: S11, obtaining image features of user behavior features, wherein the image features include pixel values, textures, and shapes; S12. Using a K-means clustering method, the data samples are divided into a number of user behavior categories according to the image features.

7. A metaverse interactive scene management system based on deep learning according to claim 6, characterized in that: The K-means clustering method comprises the following steps: Normalize the pixel values ​​of image features; Determine the appropriate number of cluster centers Q, and select Q data objects as initial cluster centers; The data objects with the smallest Euclidean distance from the initial cluster center are taken as a set to form an initial cluster; Calculate the average value of the data objects of each initial cluster as the new cluster center; The cluster center is updated repeatedly until the cluster center no longer changes, and the final cluster is determined; data objects with the same final cluster are divided into the same user behavior category.

8. According to claim 4, a metaverse interactive scene management system based on deep learning is characterized in that: In step S3, the random forest model is constructed to rank the importance of the elements of the Metaverse interaction scene, including the following steps: S31, randomly select 70% of the data objects from the data set for modeling each time; S32, randomly extracting a number of metaverse interaction scene elements as explanatory variables and user behavior categories as response variables, and generating a decision tree; S33, repeatedly generate a number of decision trees that traverse all metaverse interaction scene elements to generate a random forest; S34, calculating the prediction error of each decision tree in the random forest for the data objects not used for modeling; S35. Sort the importance of the variables of the metaverse interaction scene elements according to the prediction error of the random forest for the out-of-bag data.

9. A metaverse interactive scene management system based on deep learning according to claim 8, characterized in that: In step S33, the number of decision trees is determined by the following steps: With a step size of 5 decision trees, 20 random forests with 5 to 100 decision trees were generated; Calculate the prediction errors of 20 random forests for data objects that are not used for modeling; Select the random forest with the smallest prediction error and determine the number of decision trees.

10. The deep learning-based metaverse interactive scene management system according to claim 1, characterized in that: According to the order of importance of each element of the Metaverse Interaction Scene, the Metaverse Interaction Scene elements with an order of importance lower than n are deleted, and n is the first 50% of the Metaverse Interaction Scene elements in the order of importance.

Images