AI intelligent training platform and method supporting meta-universe multi-mode interaction

Abstract

The invention discloses an AI intelligent training platform and method supporting meta universe multi-mode interaction, and relates to the technical field of artificial intelligence, and the method comprises the following steps: converting user behavior information into multi-module interaction data; performing real-time processing and analysis on the user behavior data based on a deep learning model, performing interaction between a user and a platform, and analyzing model performance in an interaction process; predicting a user demand according to the user behavior information, automatically adjusting an interaction strategy of the platform through a continuous optimization algorithm, and obtaining user feedback in an interaction process based on a Moger model; the multi-modal interaction effect is evaluated based on fuzzy reasoning, the response mode of the platform is adjusted in time according to the multi-modal interaction effect, and continuous optimization of the platform performance is ensured. The problems of single interaction mode and lack of an effective intelligent training mechanism in the prior art are solved, and personalized and intelligent universe experience is provided for users through combination of multi-modal interaction and AI intelligent training.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology. More specifically, the present invention relates to an AI intelligent training platform and method for supporting multi-modal interaction in the metaverse. Background Art

[0002] With the continuous development of metaverse technology, more and more virtual reality and augmented reality technologies have begun to integrate into daily life, providing immersive experiences. Users in the metaverse not only need to interact with the virtual environment through traditional input methods (such as keyboards, mice, etc.), but also need to operate and control through richer multi-modal interaction methods (such as voice, gestures, eye gaze, touch, etc.).

[0003] Deficiencies of the prior art:

[0004] The introduction of AI technology has made these interactions more intelligent, capable of automatically analyzing users' behaviors, habits, and needs, and providing personalized feedback and services. However, most of the existing technologies are aimed at single interaction methods or lack effective intelligent training mechanisms, making it difficult to cope with complex and changing metaverse interaction requirements. Therefore, it is particularly important to develop an AI platform that can support multi-modal interaction in the metaverse and improve interaction efficiency through intelligent training.

[0005] In response to the above problems, the present invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an AI intelligent training platform and method for supporting multi-modal interaction in the metaverse, which introduce AI to automatically analyze users' behaviors, habits, and needs, and provide personalized feedback and services to solve the problems raised in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An AI intelligent training platform for supporting multi-modal interaction in the metaverse, comprising a multi-modal data acquisition module, a data processing module, an optimization module, and an adjustment module, with connections between the modules:

[0009] The multi-modal data acquisition module is used to sense user behavior information, identify and process the user behavior information to obtain user behavior data, and convert the user behavior information into multi-module interaction data;

[0010] The data processing module is used to perform real-time processing and analysis of the user behavior data based on a deep learning model, perform interaction between the user and the platform, and analyze the model performance during the interaction process;

[0011] An optimization module, which is used to predict user needs according to user behavior information, automatically adjust the interaction strategy of the platform through a continuous optimization algorithm, and obtain user feedback during the interaction process based on the Mogg model;

[0012] An adjustment module, which is used to evaluate the multi-modal interaction effect based on fuzzy inference according to the model performance and user feedback during the interaction process, and timely adjust the response mode of the platform according to the multi-modal interaction effect to ensure the continuous optimization of the platform performance.

[0013] In a preferred embodiment, the process of obtaining multi-module interaction data is as follows:

[0014] The behavior information includes voice input, gesture actions, and touch behaviors;

[0015] Using natural language processing technology, identify and process the user's voice input, perform noise reduction and spectrum analysis, and use a feature extraction algorithm to extract features from the voice input to obtain voice input data;

[0016] Capture visual image data through a visual sensor, denoise and filter the image data to remove environmental noise, and extract the user's gesture features of the image through a convolutional neural network deep learning method to obtain gesture feature data;

[0017] Perform signal denoising on the input of the touch screen, extract the touch behavior features of the user to obtain touch behavior data;

[0018] The user behavior data includes voice input data, gesture feature data, and touch behavior data;

[0019] Calculate the multi-module interaction data by weighted average of the voice input data, gesture feature data, and touch behavior data. The specific calculation formula is as follows:; where D is the multi-module interaction data, is the voice input data, is the weight factor of the voice input data, is the gesture feature data, is the weight factor of the gesture feature data, is the touch behavior data, and is the weight factor of the touch behavior data.

[0020] In a preferred embodiment, the specific process of the deep learning model is as follows:

[0021] Select the Transformer model to process multi-modal data according to the interaction task requirements, and train the deep learning model with a large amount of labeled data;

[0022] Use the backpropagation algorithm and the gradient descent method to optimize the parameters of the model to have a better performance for multi-modal interaction data;

[0023] Select the F1 score as the evaluation index to evaluate the performance of the model.

[0024] In a preferred embodiment, the specific process of real-time processing and analysis of the user behavior data is as follows:

[0025] Build a real-time data stream processing framework. The data stream processing framework involves multiple key components, including data collection, transmission, processing, storage, display, and monitoring;

[0026] Use Apache Kafka stream processing technology for data stream transmission, and combine with the Apache Flink stream processing engine for real-time calculation, and store the multi-module interaction data results obtained after processing and analysis into the database.

[0027] In a preferred embodiment, the steps of analyzing the model performance in the interaction process are as follows:

[0028] Calculate the absolute error between the multi-module interaction data after processing and analysis and the multi-module interaction data after weighted fusion, and compare and analyze the absolute error with a preset error threshold;

[0029] If the absolute error is less than the preset error threshold, the deep learning model analysis is accurate;

[0030] If the absolute error is greater than the preset error threshold, the deep learning model analysis is incorrect;

[0031] And calculate the accuracy rate and recall rate of the deep learning model according to the analysis results, and obtain the F1 score of the model according to the harmonic mean of the calculated accuracy rate and recall rate. The specific calculation formula is as follows:; where F1 is the F1 score of the deep learning model, P is the accuracy rate of the deep learning model, and R is the recall rate of the deep learning model.

[0032] In a preferred embodiment, the specific process of automatically adjusting the interaction strategy of the platform through continuous optimization algorithms is as follows:

[0033] Use the collaborative filtering recommendation algorithm to recommend suitable content for different users to ensure that the platform response matches the user needs;

[0034] Establish a real-time feedback mechanism, and the platform dynamically adjusts the interaction strategy according to the user's real-time feedback;

[0035] Continuously adjust and optimize the interaction strategy through A / B testing to improve the accuracy and efficiency of the platform response.

[0036] In a preferred embodiment, the specific process of obtaining user feedback in the interaction process based on the Mog model is as follows:

[0037] Determine that user satisfaction, platform efficiency, and interaction accuracy are the goals to be optimized;

[0038] Use the user feedback score as a fitness function to evaluate the effectiveness of the interaction strategy;

[0039] Generate an initial population from historical data, including different interaction strategies;

[0040] Select strategies with higher fitness;

[0041] Generate new strategies through crossover and mutation to produce the next generation;

[0042] Iteratively optimize the strategy multiple times until the optimal solution is reached or the maximum number of iterations is reached;

[0043] Quantify user satisfaction, platform efficiency, and interaction accuracy by calculating the user feedback score.

[0044] In a preferred embodiment, the specific steps for evaluating the multi-modal interaction effect based on fuzzy inference are as follows:

[0045] Define the F1 score and the user feedback score as input variables and divide them into different fuzzy sets;

[0046] Define the multi-modal interaction effect as the output variable and divide it into a fuzzy set;

[0047] Formulate fuzzy rules to describe the influence of the F1 score and the user feedback score on the multi-modal interaction effect; perform fuzzy inference according to the fuzzy rules to determine the multi-modal interaction effect.

[0048] In a preferred embodiment, the process of adjusting the platform's response mode in a timely manner according to the multi-modal interaction effect is as follows:

[0049] In multi-modal interaction, if the multi-modal interaction effect is greater than the preset multi-modal interaction threshold, the multi-modal interaction effect is good and no adjustment to the interaction process is required;

[0050] If the multi-modal interaction effect is less than the preset multi-modal interaction threshold, the multi-modal interaction effect is poor, and the platform needs to dynamically adjust the response mode according to the user's input method and context to achieve the best interaction experience and ensure the continuous optimization of the platform performance.

[0051] An AI intelligent training method for supporting multi-modal interaction in the metaverse, including the following steps:

[0052] Perceive user behavior information, identify and process the user behavior information to obtain user behavior data, and convert the user behavior information into multi-module interaction data;

[0053] Based on a deep learning model, perform real-time processing and analysis of the user behavior data, conduct interactions between the user and the platform, and analyze the model performance during the interaction process;

[0054] Predict user needs based on user behavior information, automatically adjust the platform's interaction strategy through continuous optimization algorithms, and obtain user feedback during the interaction process based on the Mog model;

[0055] According to the model performance and user feedback during the interaction process, the multimodal interaction effect is evaluated based on fuzzy reasoning, and the platform's response mode is adjusted in time according to the multimodal interaction effect to ensure continuous optimization of platform performance.

[0056] Technical effects and advantages of the AI ​​intelligent training platform and method supporting multimodal interaction of the metaverse of the present invention:

[0057] 1. The present invention supports multiple interaction modes through the platform to enhance the interactive experience between users and the virtual world. Through the AI ​​intelligent processing module, the platform can continuously analyze user behavior, continuously optimize interaction strategies, and provide personalized services; the AI ​​smart training engine can continuously improve the platform's interactive capabilities through data collection and deep learning, maintain the system's advancement and adaptability, and continuously optimize feedback mechanisms and interaction methods, so that users can enjoy a more natural, accurate, and intelligent experience.

[0058] 2. The present invention realizes a more flexible, intelligent and immersive training experience by integrating multiple modal data, overcoming the defects of the prior art of single interaction mode and insufficient intelligence. Through real-time adjustment and optimization of the AI ​​engine, the platform can provide personalized and intelligent training plans according to the user's learning progress and feedback, significantly improving the training efficiency and learning effect. The multimodal interaction function of the platform enhances the user's sense of immersion and participation, can adapt to the needs of different users, and can be widely used in various training fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a structural schematic diagram of an AI intelligent training platform that supports multimodal interaction of the metaverse according to the present invention.

[0060] Figure 2 This is a structural schematic diagram of an AI intelligent training method that supports multimodal interaction of the metaverse according to the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Embodiment 1, Figure 1This invention provides an AI intelligent training platform that supports multi-modal interaction in the metaverse.

[0063] Multi-modal data acquisition module: Used to perceive user behavior information, identify and process the user behavior information to obtain user behavior data, and convert the user behavior information into multi-module interaction data recognizable by the platform;

[0064] The behavior information includes voice input, gesture actions, and touch behaviors;

[0065] The user behavior data includes voice input data, gesture feature data, and touch behavior data;

[0066] Voice input recognition: Using natural language processing technology, identify and process the user's voice input, perform noise reduction and spectrum analysis, and use a feature extraction algorithm (such as MFCC) to extract features from the voice input to obtain voice input data;

[0067] Gesture action recognition: Capture visual image data through a visual sensor (such as a camera), denoise and filter the image data to remove environmental noise, and extract the user's gesture features from the image through a deep learning method of convolutional neural network (CNN) to obtain gesture feature data to represent the user's gesture actions and achieve non-contact interaction;

[0068] Touch behavior recognition: Denoise the signal of the touch screen input, extract the user's touch behavior features to obtain touch behavior data, and enhance the interaction experience;

[0069] The voice input data, gesture feature data, and touch behavior data are calculated through weighted average to obtain multi-module interaction data. The specific calculation formula is as follows: D = w 1 f 1 + w 2 f 2 + w 3 f 3 ; where D is the multi-module interaction data, f 1 is the voice input data, w 1 is the weight factor of the voice input data, f 2 is the gesture feature data, w 2 is the weight factor of the gesture feature data, f 3 is the touch behavior data, w 3 is the weight factor of the touch behavior data;

[0070] Data processing module: Used to perform real-time processing and analysis on the user behavior data based on a deep learning model, and conduct interaction between the user and the platform, and analyze the model performance during the interaction process;

[0071] The specific process of implementing the deep learning model is as follows:

[0072] Select a Transformer model to process multimodal data according to the requirements of interaction tasks, and train the deep learning model with a large amount of labeled data;

[0073] Use the backpropagation algorithm and the gradient descent method to optimize the parameters of the model to ensure its good performance on multimodal interaction data;

[0074] Select the F1 score as the evaluation metric to evaluate the performance of the model;

[0075] The specific process of real-time processing and analysis of the user behavior data is as follows:

[0076] Build a real-time data stream processing framework. The data stream processing framework involves multiple key components, including data collection, transmission, processing, storage, display, and monitoring;

[0077] Use Apache Kafka stream processing technology for data stream transmission, and combine it with the Apache Flink stream processing engine for real-time calculation. Finally, store the processed and analyzed multi-module interaction data results in a database;

[0078] Compare and analyze the processed and analyzed multi-module interaction data with the weighted-fused multi-module interaction data. Compare the absolute error between the processed and analyzed multi-module interaction data and the weighted-fused multi-module interaction data with a preset error threshold. If the absolute error is less than the preset error threshold, the deep learning model analysis is accurate;

[0079] If the absolute error is greater than the preset error threshold, the deep learning model analysis is incorrect;

[0080] And calculate the accuracy and recall rate of the deep learning model according to the analysis results, and obtain the F1 score of the model according to the harmonic mean of the calculated accuracy and recall rate; the specific calculation formula is as follows: In the formula, F1 is the F1 score of the deep learning model, P is the accuracy of the deep learning model, and R is the recall rate of the deep learning model;

[0081] Selecting the F1 score as the evaluation metric to evaluate the performance of the model has the following effects:

[0082] Balance precision and recall: The F1 score combines precision and recall, and is especially suitable for dealing with the situation of imbalanced datasets;

[0083] Model evaluation: Using precision or recall alone may not fully reflect the performance of the model. The F1 score provides a comprehensive metric considering the classification effect of the model, making the evaluation more comprehensive;

[0084] Applicable to tasks with class imbalance: The F1 score is particularly suitable for situations with class imbalance. When the number of positive class samples in the dataset is much less than that of negative class samples, the F1 score can better reflect the performance of the model in predicting rare events, and is not as susceptible to bias from the larger class as precision or recall;

[0085] Model tuning: The F1 score can be used as an objective function in the hyperparameter tuning process to help select the optimal model hyperparameter settings, especially in situations where precision and recall need to be balanced;

[0086] Optimization module: Predict user needs based on user behavior information, automatically adjust the platform's interaction strategy through a continuous optimization algorithm, and obtain user feedback during the interaction based on the MOG model;

[0087] Accurately identify the current demand type and intention of the user based on data analysis of user behavior information;

[0088] The specific process of automatically adjusting the platform's interaction strategy through the continuous optimization algorithm is as follows:

[0089] Use collaborative filtering recommendation algorithms to recommend suitable content or services for different users to ensure that the platform response matches the user needs;

[0090] Establish a real-time feedback mechanism, and the platform dynamically adjusts the interaction strategy according to the user's real-time feedback;

[0091] Continuously adjust and optimize the interaction strategy through A / B testing to improve the accuracy and efficiency of the platform response.

[0092] The MOG model (MOG, Multi-Objective Genetic Algorithm) is a multi-objective genetic algorithm that can optimize multiple objective functions (such as user satisfaction, platform response speed, resource consumption, etc.). When analyzing user feedback, MOG can help the platform find a balance among multiple objectives and optimize strategies.

[0093] The specific process of obtaining user feedback during the interaction based on the MOG model is as follows:

[0094] Determine that user satisfaction, platform efficiency, and interaction accuracy are the objectives to be optimized;

[0095] Use the user feedback score as a fitness function to evaluate the effect of the interaction strategy;

[0096] Generate an initial population from historical data, including different interaction strategies;

[0097] Select strategies with higher fitness;

[0098] Generate new strategies through crossover and mutation to produce the next generation;

[0099] Iteratively optimize the strategy multiple times until the optimal solution is reached or the maximum number of iterations is reached;

[0100] Quantify user satisfaction, platform efficiency, and interaction accuracy by calculating the user feedback score.

[0101] Adjustment module: Based on the model interaction performance and user feedback during the interaction process, evaluate the multimodal interaction effect based on fuzzy inference, and adjust the platform's response mode in a timely manner according to the multimodal interaction effect to ensure continuous optimization of the platform performance.

[0102] The specific steps for evaluating the multimodal interaction effect based on fuzzy inference are as follows:

[0103] Step C1, Define the F1 score and the user feedback score as input variables and divide them into different fuzzy sets respectively;

[0104] For example, "Low", "Medium", "High" for the F1 score, and "Low", "Medium", "High" for the user feedback score;

[0105] Step C2, Define the multimodal interaction effect as the output variable and divide it into a fuzzy set. For example, "Low", "High" for the multimodal interaction effect;

[0106] Step C3, Develop a set of fuzzy rules to describe the influence of different input variables on the output variable. The definition of the rules can be based on professional knowledge or obtained through data analysis and experiments. For example:

[0107] Mark the F1 score as F1, the user feedback score as Y, and the multimodal interaction effect as P, then the following can be defined

[0108] Rule 1: If (F1 is High) and (Y is High), then (P is High)

[0109] Rule 2: If (F1 is Low) and (Y is Low), then (P is Low) ...

[0111] Step C4, Perform fuzzy inference according to the fuzzy rules to determine the multimodal interaction effect.

[0112] It should be noted that the division of the fuzzy sets can be adjusted according to the actual situation. For example, although this embodiment takes three fuzzy sets as an example, in fact, the F1 score, the user feedback score, and the multimodal interaction effect can be divided into more than three sets to facilitate better accurate recognition.

[0113] Furthermore, for the F1 score, the judgment of the user feedback rating can be determined by setting a threshold according to the actual situation; when the F1 score is higher than 0.7, it is labeled as "High", when the user feedback rating is higher than 90 points (rating range 0 - 100 points), it is labeled as "High", etc., which will not be elaborated here.

[0114] In multimodal interaction, if P is High, the multimodal interaction effect is greater than the preset multimodal interaction threshold, and the multimodal interaction effect is good, and there is no need to adjust the interaction process;

[0115] If P is Low, the multimodal interaction effect is less than the preset multimodal interaction threshold, and the multimodal interaction effect is not good. The platform needs to dynamically adjust the response mode according to the user's input method and context to achieve the best interaction experience and ensure the continuous optimization of the platform performance.

[0116] The platform should flexibly switch the response mode according to the user's input method. For example:

[0117] If the user uses voice input, the platform should preferentially select voice output for response, so that the user can avoid looking back at the screen.

[0118] When the user operates through the touch screen, the platform can provide instant visual feedback or tactile feedback to increase the immediacy of the interaction.

[0119] For long texts or complex operations, the platform can provide step-by-step guidance to improve user understanding through a combination of text and voice.

[0120] Example 2, Figure 2 provides an AI intelligent training method for supporting multimodal interaction in the metaverse.

[0121] Perceive user behavior information, identify and process the user behavior information to obtain user behavior data, and convert the user behavior information into multi-module interaction data;

[0122] Based on the deep learning model, perform real-time processing and analysis on the user behavior data, conduct interaction between the user and the platform, and analyze the model performance during the interaction process;

[0123] Predict user needs according to user behavior information, automatically adjust the interaction strategy of the platform through continuous optimization algorithms, and obtain user feedback during the interaction process based on the Mog model;

[0124] Based on the model performance and user feedback during the interaction process, evaluate the multimodal interaction effect based on fuzzy inference, and adjust the response mode of the platform in a timely manner according to the multimodal interaction effect to ensure the continuous optimization of the platform performance.

[0125] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0126] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0127] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0128] In addition, in each embodiment of this application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0129] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0130] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An AI intelligent training platform supporting multimodal interaction of the metaverse, characterized in that: It includes multimodal data acquisition module, data processing module, optimization module and adjustment module. There are connections between modules: A multimodal data acquisition module is used to sense user behavior information, identify and process the user behavior information, obtain user behavior data, and convert the user behavior data into multi-module interaction data; A data processing module, used to process and analyze the user behavior data in real time based on a deep learning model, and to perform interaction between the user and the platform, and to analyze the model performance during the interaction process; The optimization module is used to predict user needs based on user behavior information, automatically adjust the platform's interaction strategy through continuous optimization algorithms, and obtain user feedback during the interaction process based on the Mog model; The adjustment module is used to evaluate the multimodal interaction effect based on fuzzy reasoning according to the model performance and user feedback during the interaction process, and to adjust the platform's response mode in a timely manner according to the multimodal interaction effect to ensure continuous optimization of the platform performance.

2. According to claim 1, an AI intelligent training platform supporting multimodal interaction of the metaverse is characterized in that: The multi-module interaction data acquisition process is as follows: The behavior information includes voice input, gesture action, and touch behavior; Use natural language processing technology to recognize and process the user's voice input, perform noise reduction and spectrum analysis, and use feature extraction algorithms to extract features from the voice input to obtain voice input data; Capture visual image data through visual sensors, denoise and filter the image data to remove environmental noise, and extract user gesture features from the image through a convolutional neural network deep learning method to obtain gesture feature data; Perform signal denoising on the touch screen input, extract the user's touch behavior characteristics, and obtain touch behavior data; The user behavior data includes voice input data, gesture feature data, and touch behavior data; The multi-module interaction data is obtained by weighted averaging the voice input data, gesture feature data, and touch behavior data. The specific calculation formula is as follows: D=w1f1+w2f2+w3f3; where D is the multi-module interaction data, f1 is the voice input data, w1 is the voice input data weight factor, f2 is the gesture feature data, w2 is the gesture feature data weight factor, f3 is the touch behavior data, and w3 is the touch behavior data weight factor.

3. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 2, characterized in that: The specific process of implementing the deep learning model is as follows: Select the Transformer model to process multimodal data according to the interactive task requirements, and train the deep learning model through large-scale labeled data; Use back-propagation and gradient descent to optimize the model's parameters to achieve better performance on multimodal interaction data; The F1 score is selected as the evaluation metric to assess the performance of the model.

4. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 3, characterized in that: The specific process of real-time processing and analysis of the user behavior data is as follows: Build a real-time data stream processing framework involving multiple key components, including data collection, transmission, processing, storage, display, and monitoring; By using Apache Kafka stream processing technology for data stream transmission and combining it with the Apache Flink stream processing engine for real-time computing, the multi-module interaction data results obtained after processing and analysis are stored in the database.

5. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 4, characterized in that: The steps for analyzing the model performance in the interactive process are as follows: Calculate the absolute error between the multi-module interaction data after processing and analysis and the multi-module interaction data after weighted fusion, and compare and analyze the absolute error with a preset error threshold; If the absolute error is less than the preset error threshold, the deep learning model analysis is accurate; If the absolute error is greater than the preset error threshold, the deep learning model analysis is wrong; The accuracy and recall of the deep learning model are calculated based on the analysis results, and the F1 score of the model is obtained by calculating the harmonic mean of the accuracy and recall. The specific calculation formula is as follows: Where F1 is the F1 score of the deep learning model, P is the precision of the deep learning model, and R is the recall of the deep learning model.

6. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 5, characterized in that: The specific process of automatically adjusting the platform's interaction strategy through continuous optimization algorithm is as follows: Use collaborative filtering recommendation algorithms to recommend suitable content to different users, ensuring that platform responses match user needs; Establish a real-time feedback mechanism, and the platform dynamically adjusts the interaction strategy based on users' real-time feedback; Continuously adjust and optimize interaction strategies through A / B testing to improve the accuracy and efficiency of platform responses.

7. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 6, characterized in that: The specific process of obtaining user feedback in the interaction process based on the Mog model is as follows: Identify user satisfaction, platform efficiency, and interaction accuracy as goals that need to be optimized; Use user feedback scores as fitness functions to evaluate the effectiveness of interaction strategies; Generate an initial population from historical data, including different interaction strategies; Choose a strategy with higher fitness; Generate new strategies through crossover and mutation to generate the next generation; Iterate the optimization strategy multiple times until the optimal solution is reached or the maximum number of iterations is reached; Quantify user satisfaction, platform efficiency, and interaction accuracy by calculating user feedback scores.

8. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 7, characterized in that: The specific steps of evaluating the multimodal interaction effect based on fuzzy reasoning are as follows: Define the F1 score and user feedback score as input variables and divide them into different fuzzy sets respectively; Define the multimodal interaction effect as the output variable and divide it into fuzzy sets; Formulate fuzzy rules to describe the impact of F1 score and user feedback rating on multimodal interaction effects; Fuzzy reasoning is performed based on fuzzy rules to determine the multimodal interaction effect.

9. The AI ​​intelligent training platform supporting multimodal interaction of the metaverse according to claim 8, characterized in that: In the process of timely adjusting the response mode of the platform according to the multimodal interaction effect: In multimodal interaction, if the multimodal interaction effect is greater than a preset multimodal interaction threshold, the multimodal interaction effect is good and there is no need to adjust the interaction process; If the multimodal interaction effect is greater than the preset multimodal interaction threshold, the multimodal interaction effect is not good. The platform needs to dynamically adjust the response mode according to the user's input method and context to achieve the best interactive experience and ensure continuous optimization of platform performance.

10. An AI intelligent training method supporting multimodal interaction of the metaverse, characterized in that: The following steps are involved: Perceiving user behavior information, identifying and processing the user behavior information to obtain user behavior data, and converting the user behavior data into multi-module interaction data; Process and analyze the user behavior data in real time based on a deep learning model, conduct interaction between the user and the platform, and analyze the model performance during the interaction process; Predict user needs based on user behavior information, automatically adjust the platform's interaction strategy through continuous optimization algorithms, and obtain user feedback during the interaction process based on the Mog model; According to the model performance and user feedback during the interaction process, the multimodal interaction effect is evaluated based on fuzzy reasoning, and the platform's response mode is adjusted in time according to the multimodal interaction effect to ensure continuous optimization of platform performance.

Images