A user interaction experience optimization method and system based on metaverse

Abstract

The present invention discloses a user interaction experience optimization method and system based on metaverse, and relates to the field of metaverse interaction technology. A user interaction experience optimization system based on metaverse, including: an interaction feature extraction module and an interaction reward experience optimization module. The present invention enhances the in-depth understanding of user interaction behavior through multimodal feature coupling; improves the accuracy of visual interaction information processing by using an improved convolutional neural network model; improves the analysis quality of auditory and text interaction information through distributed discrete cosine transform and text semantic representation; the user interaction reward analysis model realizes accurate evaluation of the value of user interaction behavior and dynamic update of reward value-added through improvement of reward value quantification and group optimization algorithm; can more effectively motivate and guide user behavior, optimize user experience, and provide strong data support for user behavior analysis and decision-making.

Description

Technical Field

[0001] The present invention relates to the field of metaverse interaction technology, and in particular to a metaverse-based user interaction experience optimization method and system. Background Art

[0002] The metaverse is a digital space supported by technologies such as virtual reality, augmented reality and artificial intelligence. By simulating the real world or building a virtual world, it allows users to interact, socialize, create, trade and entertain. However, in the process of user interaction, the frequency of interaction is often not active enough. Therefore, it is particularly important to establish an interactive experience reward mechanism to optimize the user's interactive experience. The reward mechanism based on user behavior characteristics can provide users with personalized and precise incentives by analyzing the depth, frequency and contribution of interaction. This can not only stimulate users' willingness to interact and enhance their loyalty and participation, but also create richer data for the platform and build an active community ecology. Summary of the invention

[0003] The present invention aims to provide a method and system for optimizing user interaction experience based on the metaverse to improve user interaction enthusiasm.

[0004] A method for optimizing user interaction experience based on Metaverse, comprising the following steps:

[0005] Obtain user virtual interaction information; the user virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information; input the user virtual interaction information into the user interaction feature extraction model for feature extraction to obtain a user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modal feature recognition layer, an auditory modal feature recognition layer, a text modal feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and in the user interaction feature extraction model, a multimodal feature analysis method is used to improve and analyze the convolutional neural network model;

[0006] Based on the user virtual interaction information feature set and the user interaction reward analysis model, feature recognition is performed on the user virtual interaction information feature set to obtain the target user interaction reward increment; the user social interaction information is updated based on the target user interaction reward increment every preset time range;

[0007] The user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer. A reward value quantization function is deployed in the reward value analysis layer. In the reward value analysis layer, the group optimization algorithm is improved and updated to construct a reward value quantization function.

[0008] As a preferred technical solution of the present invention, the user interaction feature extraction model includes a data classification layer, a visual modality feature recognition layer, an auditory modality feature recognition layer, a text modality feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer;

[0009] The data classification layer is used to input the user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information in the user virtual interaction information into the corresponding feature recognition layer for analysis;

[0010] Among them, the user visual interaction information is input into the visual modality feature recognition layer, the user auditory interaction information is input into the auditory modality feature recognition layer, the user text interaction information is input into the text modality feature recognition layer, and the user social interaction information is input into the social information feature recognition layer;

[0011] The visual modality feature recognition layer is used to extract multiple features of user visual interaction information to obtain user visual interaction fusion features;

[0012] The auditory modality feature recognition layer is used to extract Mel-frequency cepstral coefficients from the user auditory interaction information using distributed discrete cosine transform to obtain the user auditory interaction features;

[0013] The text modality feature recognition layer is used to extract text semantic representation from user text interaction information and obtain user text interaction features;

[0014] The social information feature recognition layer is used to extract features of user social interaction information and obtain user social interaction features;

[0015] The multimodal feature coupling layer is used to couple the user visual interaction fusion features, user auditory interaction features, user text interaction features and user social interaction features using Bayesian expert product to obtain the user virtual interaction information feature set.

[0016] As a preferred technical solution of the present invention, the specific steps of performing multi-feature extraction in the visual modality feature recognition layer include:

[0017] Three feature extraction convolution blocks and a feature fusion module are constructed in the visual modality feature recognition layer; the first dense convolution block is used to extract the global features of the user visual interaction information to obtain the global features of the user visual interaction; the second dense convolution block is used to extract the emotional features of the user visual interaction information to obtain the emotional features of the user visual interaction; the third dense convolution block is used to extract the action features of the user visual interaction information to obtain the action features of the user visual interaction;

[0018] The specific steps of constructing any one of the three feature extraction convolution blocks include:

[0019] A convolutional layer is constructed using a single multi-box detector, and multiple convolutional layers are connected to construct a feature extraction convolutional block; the corresponding sub-features are output in each convolutional layer, and the output features of the previous convolutional layer are used as the input features of the next convolutional layer. After all sub-features are average-pooled in the last convolutional layer, the output features of the corresponding feature extraction convolutional block are obtained;

[0020] In the feature fusion module, the user visual interaction global features, the user visual interaction emotional features and the user visual interaction action features are fused to obtain the user visual interaction fusion features.

[0021] As a preferred technical solution of the present invention, the specific steps of extracting text semantic representation from user text interaction information in the text modality feature recognition layer include:

[0022] A pre-trained semantic analysis model is constructed, and a pre-trained sentiment recognition model is connected to the pre-trained semantic analysis model to form a text modal feature recognition layer; in the pre-trained semantic analysis model, semantic vector analysis is performed on user text interaction information to obtain a number of user text interaction vectors; in the pre-trained sentiment recognition model, sentiment recognition is performed on each user text interaction vector, and sentiment labels are added to obtain a number of user text interaction vectors with sentiment labels; all user text interaction vectors with sentiment labels are connected to obtain user text interaction features.

[0023] As a preferred technical solution of the present invention, the user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer;

[0024] The data preprocessing layer is used to preprocess the user virtual interaction information feature set to obtain the preprocessed virtual interaction information feature value T n , n=1, 2, …, N; N is the total number of features in the user virtual interaction information feature set;

[0025] The reward value analysis layer is used to calculate the reward value based on all preprocessed virtual interaction information feature values ​​T n Analyze with the reward value quantization function to obtain the target user interaction reward increment;

[0026] The result output layer is used to output the target user interaction reward increment.

[0027] As a preferred technical solution of the present invention, the specific steps of constructing the reward value quantization function in the reward value regression analysis layer include:

[0028] Collect feature recognition reward increment data sets; the feature recognition reward increment data sets contain several sets of feature value sets with target value annotations, and the target value corresponds to the verified reward increment;

[0029] Construct K simulated reward value quantization function individuals H based on feature recognition reward increment dataset k ; Construct the initial individual position matrix; The initial individual position matrix is ​​used to represent each simulated reward value quantization function individual H k The initial position of; K simulated reward value quantization function individual H k Combine to obtain the reward value quantization function iterative population; set the maximum number of iterations;

[0030] The specific steps of calculating the fitness of the individual of the simulated reward value quantization function are: using the individual of the simulated reward value quantization function to calculate the feature recognition reward increment data set to obtain the simulated feature increment; calculating the Euclidean distance between the simulated feature increment and the target value in the feature recognition reward increment data set to obtain the simulated calculation difference; and taking the inverse of the simulated calculation difference as the fitness of the individual of the simulated reward value quantization function;

[0031] Set the position update formula F(t+1), F(t+1)=F(t)+ Q; where t represents the current iteration number, and F(t) represents the initial individual position matrix when the current iteration number is t; Q represents random update of individual positions;

[0032] Based on the position update formula F(t+1), the initial individual position matrix is ​​updated to obtain a new initial individual position matrix; based on the new initial individual position matrix, the reward value quantization function iteration population is iterated to obtain a new reward value quantization function iteration population;

[0033] When the maximum number of iterations is reached, the simulated reward value quantization function individual corresponding to the maximum output fitness is the optimal simulated reward value quantization function individual; the reward value quantization function in the reward value regression analysis layer is constructed based on the optimal simulated reward value quantization function individual.

[0034] As a preferred technical solution of the present invention, the specific steps of updating the user social interaction information based on the target user interaction reward increment every preset time range include:

[0035] Calculate the total amount of all target user interaction reward appreciation within the preset time range to obtain the stage target user interaction reward appreciation; match the stage target user interaction reward appreciation in the user interaction implicit value matching library to obtain the corresponding updated target user interaction level; update the user social interaction information based on the updated target user interaction level to obtain new user social interaction information.

[0036] A user interaction experience optimization system based on Metaverse, comprising:

[0037] The interactive feature extraction module includes a data acquisition unit and a feature analysis unit; the data acquisition unit is used to acquire user virtual interaction information; the user virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information; the feature analysis unit is used to input the user virtual interaction information into the user interaction feature extraction model for feature extraction to obtain a user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modal feature recognition layer, an auditory modal feature recognition layer, a text modal feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and the multimodal feature analysis method is used in the user interaction feature extraction model to improve and analyze the convolutional neural network model;

[0038] The interactive reward experience optimization module includes a reward analysis unit and an information optimization unit; the reward analysis unit is used to perform feature recognition on the user virtual interaction information feature set based on the user virtual interaction information feature set and the user interaction reward analysis model to obtain the target user interaction reward value-added; the user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer, a reward value quantization function is deployed in the reward value analysis layer, the group optimization algorithm is improved and updated in the reward value analysis layer, and a reward value quantization function is constructed; the information optimization unit is used to update the user social interaction information based on the target user interaction reward value-added every preset time range.

[0039] The present invention has the following advantages:

[0040] 1. The present invention enhances the in-depth understanding of user interaction behavior through multimodal feature coupling; improves the accuracy of visual interaction information processing by using an improved convolutional neural network model; improves the analysis quality of auditory and text interaction information through distributed discrete cosine transform and text semantic representation; the social information feature recognition layer enhances the ability to capture social interaction dynamics; the user interaction reward analysis model realizes accurate evaluation of the value of user interaction behavior and dynamic update of reward value-added through improvement of reward value quantification and group optimization algorithm; can more effectively motivate and guide user behavior, optimize user experience, and provide strong data support for user behavior analysis and decision-making.

[0041] 2. The present invention realizes the deep fusion and analysis of user vision, hearing, text and social interaction information through multimodal feature analysis method, improves the accuracy of feature extraction, and enhances the comprehensiveness of information through multimodal feature coupling; utilizes the improved convolutional neural network model to improve the multi-feature extraction capability of visual interaction information; extracts the Mel-frequency cepstral coefficients of auditory interaction information through distributed discrete cosine transform, and enhances the processing of auditory information; improves the understanding of text interaction information through text semantic representation extraction; and at the same time, the extraction of social information features enhances the understanding of social interaction dynamics, and can more accurately capture and analyze the user's interactive behavior in the virtual environment, providing strong data support for subsequent user behavior analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the structure of a user interaction experience optimization system based on the metaverse adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0044] Embodiment 1, a method for optimizing user interaction experience based on Metaverse, comprising the following steps:

[0045] Acquire user virtual interaction information; the user virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information;

[0046] In the context of the metaverse, users’ virtual interaction information is not limited to traditional physical world interactions, but spans multiple virtual environments and multimodal information interactions. Generally speaking, users interact with other users, environments, and virtual objects through virtual avatars. In such an environment, users’ interaction information becomes richer and more complex, and includes multimodal data such as visual, auditory, textual, and social interaction information, which reflects users’ behaviors, emotions, intentions, and social activities in the virtual world. User virtual interaction information refers to various behaviors, perceptions, and interaction data of users in the metaverse, which are reflected through multiple sensory inputs and social interaction forms. It includes not only users’ input information, but also the virtual environment’s response to users, reflecting users’ immersive experience and social behaviors in the virtual world.

[0047] When users interact, accurate feature analysis can be performed on each interaction to assign implicit reward values. Users can enjoy value-added services based on the reward values, thereby improving their interactive experience in the context of the Metaverse.

[0048] The user virtual interaction information is input into the user interaction feature extraction model for feature extraction to obtain the user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modality feature recognition layer, an auditory modality feature recognition layer, a text modality feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and the multimodal feature analysis method is used in the user interaction feature extraction model to improve and analyze the convolutional neural network model;

[0049] The data classification layer is used to input the user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information in the user virtual interaction information into the corresponding feature recognition layer for analysis;

[0050] Among them, the user visual interaction information is input into the visual modality feature recognition layer, the user auditory interaction information is input into the auditory modality feature recognition layer, the user text interaction information is input into the text modality feature recognition layer, and the user social interaction information is input into the social information feature recognition layer;

[0051] The visual modality feature recognition layer is used to extract multiple features of user visual interaction information to obtain user visual interaction fusion features;

[0052] The auditory modal feature recognition layer is used to extract Mel-frequency cepstral coefficients from the user's auditory interaction information using distributed discrete cosine transform to obtain the user's auditory interaction features. Using distributed discrete cosine transform to extract Mel-frequency cepstral coefficients to analyze the user's auditory interaction information can effectively capture key information such as speech and emotion in the audio data and enhance the model's understanding of sound. Through these efficient feature extraction methods, the model can more accurately understand and represent the user's interaction features.

[0053] The text modality feature recognition layer is used to extract text semantic representation from user text interaction information and obtain user text interaction features;

[0054] The social information feature recognition layer is used to extract features from user social interaction information to obtain user social interaction features; user social interaction information is the user's implicit interaction level, indicating the user's enthusiasm and participation in interaction;

[0055] The multimodal feature coupling layer is used to couple the user visual interaction fusion features, user auditory interaction features, user text interaction features and user social interaction features using the Bayesian expert product to obtain the user virtual interaction information feature set; in the multimodal feature coupling layer, the Bayesian expert product method is used to fuse the information of each modality. This method can effectively handle the different features and uncertainties between modalities. The Bayesian expert product allows the features of each modality to be combined in a weighted manner, making the final feature set more representative and information-dense, which helps to make up for the limitations of single modality features and enhance the complementarity between different modalities, thereby improving the adaptability and robustness of the model in multiple tasks and multiple scenarios;

[0056] By using independent feature recognition layers of four modalities, namely vision, hearing, text, and social information, in the model, combined with a multimodal feature coupling layer, different types of information can be effectively integrated. The integration of multimodal data enables the model to fully capture the multidimensional characteristics of user behavior and improve the ability to understand and analyze user interactions. For example, the combination of vision and hearing can better describe the user's immersive experience in the virtual world, while the combination of social information and text helps analyze the user's social behavior.

[0057] The specific steps of performing multi-feature extraction in the visual modality feature recognition layer include:

[0058] Three feature extraction convolution blocks and a feature fusion module are constructed in the visual modality feature recognition layer; the first dense convolution block is used to extract the global features of the user visual interaction information to obtain the global features of the user visual interaction; the second dense convolution block is used to extract the emotional features of the user visual interaction information to obtain the emotional features of the user visual interaction; the third dense convolution block is used to extract the action features of the user visual interaction information to obtain the action features of the user visual interaction;

[0059] The specific steps of constructing any one of the three feature extraction convolution blocks include:

[0060] A convolutional layer is constructed using a single multi-box detector, and multiple convolutional layers are connected to construct a feature extraction convolutional block; the corresponding sub-features are output in each convolutional layer, and the output features of the previous convolutional layer are used as the input features of the next convolutional layer. After all sub-features are average-pooled in the last convolutional layer, the output features of the corresponding feature extraction convolutional block are obtained;

[0061] In the feature fusion module, the user visual interaction global features, the user visual interaction emotional features and the user visual interaction action features are fused to obtain the user visual interaction fusion features;

[0062] Global features capture the large-scale scene and environmental information in the user's visual interaction, which helps to understand the user's overall behavior pattern in the virtual world; emotional features capture the user's emotional expression in visual interaction, such as facial expressions, posture changes, etc., which is crucial for identifying the user's emotional state and psychological reaction; action features focus on the user's movement and behavior patterns in the virtual environment, such as walking, running, waving, etc., which helps to analyze the user's dynamic behavior;

[0063] Each convolution block consists of multiple convolution layers, each of which gradually extracts more abstract and advanced features. The convolution layers are constructed by a single multi-box detector and connected to form a dense convolution block, which can effectively extract more fine-grained visual features and avoid feature loss and information redundancy through layered processing. The multi-box detector can extract features at multiple locations in the same layer, and through parallel processing of multiple convolution kernels, the model's perception of complex image content is enhanced, especially when dealing with multiple objects, different scenes and user interactions.

[0064] Through the feature fusion module, the global features, emotional features and action features extracted from different convolution blocks are fused, and the model can obtain more comprehensive visual interaction features; fusing features from different convolution blocks helps capture the multi-dimensional relationship of visual information; for example, the user's emotional state may affect his action pattern, and the global environment features can provide important contextual information. Feature fusion helps to form a richer feature vector; by fusing different types of features, the model can make up for the lack of information that a single feature may bring; for example, relying solely on action features may not be enough to judge the user's emotions, while emotional features can provide emotional background for action features, so that the model can understand the user's behavior more comprehensively and accurately;

[0065] The specific steps of extracting text semantic representation from user text interaction information in the text modality feature recognition layer include:

[0066] Construct a pre-trained semantic analysis model, and connect the pre-trained emotion recognition model to the pre-trained semantic analysis model to form a text modal feature recognition layer; in the pre-trained semantic analysis model, perform semantic vector analysis on user text interaction information to obtain several user text interaction vectors; in the pre-trained emotion recognition model, perform emotion recognition on each user text interaction vector, add emotion tags, and obtain several user text interaction vectors with emotion tags; connect all user text interaction vectors with emotion tags to obtain user text interaction features;

[0067] The pre-trained semantic analysis model is trained based on the bag-of-words model, and the pre-trained sentiment recognition model is trained based on the BP neural network model;

[0068] By building a pre-trained semantic analysis model, the model can perform in-depth semantic understanding and analysis of user text interaction information to obtain high-quality text semantic vectors; the pre-trained semantic analysis model can fully capture the contextual information in the text, making the semantic representation of the text more accurate, and can understand the relevance between words, grammatical structure and the intention behind it; after semantic analysis, the text interaction information is converted into semantic vectors, which contain rich semantic features of the text and are helpful for subsequent tasks such as sentiment analysis and behavior prediction; sentiment analysis enables the model to not only understand the basic semantic information of the text, but also capture the emotional state expressed by the user in the interaction; through the sentiment recognition model, the model can add specific sentiment labels to each text interaction vector, such as pleasure, anger, sadness, etc., which not only helps to understand the user's emotional state, but also provides more accurate sentiment dimension data for tasks such as personalized recommendation and user behavior analysis;

[0069] Based on the user virtual interaction information feature set and the user interaction reward analysis model, feature recognition is performed on the user virtual interaction information feature set to obtain the target user interaction reward increment; the user social interaction information is updated based on the target user interaction reward increment every preset time range;

[0070] The user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer, and a result output layer. A reward value quantization function is deployed in the reward value analysis layer. In the reward value analysis layer, the group optimization algorithm is improved and updated to construct a reward value quantization function.

[0071] The data preprocessing layer is used to preprocess the user virtual interaction information feature set to obtain the preprocessed virtual interaction information feature value T n , n=1, 2, …, N; N is the total number of features in the user virtual interaction information feature set;

[0072] The reward value analysis layer is used to calculate the reward value based on all preprocessed virtual interaction information feature values ​​T n Analyze with the reward value quantization function to obtain the target user interaction reward increment;

[0073] The result output layer is used to output the target user interaction reward increment;

[0074] The specific steps to construct the reward value quantization function in the reward value regression analysis layer include:

[0075] Collect feature recognition reward increment data sets; the feature recognition reward increment data sets contain several sets of feature value sets with target value annotations, and the target value corresponds to the verified reward increment;

[0076] Construct K simulated reward value quantization function individuals H based on feature recognition reward increment dataset k; Construct the initial individual position matrix; The initial individual position matrix is ​​used to represent each simulated reward value quantization function individual H k The initial position of; K simulated reward value quantization function individual H k Combination, to obtain the iterative population of the reward value quantization function; set the maximum number of iterations; the maximum number of iterations is set by professional technicians according to actual conditions;

[0077] The specific steps of calculating the fitness of the individual of the simulated reward value quantization function are: using the individual of the simulated reward value quantization function to calculate the feature recognition reward increment data set to obtain the simulated feature increment; calculating the Euclidean distance between the simulated feature increment and the target value in the feature recognition reward increment data set to obtain the simulated calculation difference; and taking the inverse of the simulated calculation difference as the fitness of the individual of the simulated reward value quantization function;

[0078] Set the position update formula F(t+1), F(t+1)=F(t)+ Q; where t represents the current iteration number, and F(t) represents the initial individual position matrix when the current iteration number is t; Q represents random update of individual positions;

[0079] Based on the position update formula F(t+1), the initial individual position matrix is ​​updated to obtain a new initial individual position matrix; based on the new initial individual position matrix, the reward value quantization function iteration population is iterated to obtain a new reward value quantization function iteration population;

[0080] When the maximum number of iterations is reached, the simulated reward value quantization function individual corresponding to the maximum fitness is output, which is the optimal simulated reward value quantization function individual; the reward value quantization function in the reward value regression analysis layer is constructed based on the optimal simulated reward value quantization function individual;

[0081] By constructing a reward value quantization function, the model can accurately establish a mathematical relationship between the user's virtual interaction information feature value and the actual reward increment; the optimization of the reward value quantization function makes the reward value-added analysis more accurate and can better reflect the corresponding relationship between user interaction behavior and rewards; the reward value quantization function can learn to extract key factors from user behavior characteristics and accurately predict the reward value-added by training and optimizing a large number of feature value sets labeled with target values; using the position update formula, the individual search position can be dynamically updated, so that the group optimization algorithm can adaptively adjust the search strategy and improve the search efficiency; randomly updating the individual position can enable the model to maintain a certain flexibility during the search process, avoid being restricted by a fixed mode, and improve the algorithm's exploration ability; by preprocessing the feature set of the user's virtual interaction information and using the improved reward value quantization function to analyze the user's interaction reward value-added, personalized reward analysis can be achieved; the virtual interaction information characteristics of different users may vary greatly, and the reward value quantization function can tailor the calculation method of reward value-added according to each user's interaction mode and feature data to ensure that each user's reward analysis is more in line with their actual behavior;

[0082] The specific steps of updating the user social interaction information based on the target user interaction reward increment every preset time range include:

[0083] Calculate the total amount of all target user interaction reward increments within a preset time range to obtain the stage target user interaction reward increments; match the stage target user interaction reward increments in the user interaction implicit value matching library to obtain the corresponding updated target user interaction level; update the user social interaction information based on the updated target user interaction level to obtain new user social interaction information;

[0084] The preset time range is set by professional technicians based on actual conditions; by regularly calculating and updating the value-added of user interaction rewards, the system can timely adjust the user's social status, making the weight on the social platform more fair and dynamic. This adjustment can reflect the user's status in the social network in real time based on the user's latest behavior and contribution, thereby avoiding inaccurate social rankings caused by the lag of the old rating system or behavior patterns; the real-time update of the value-added of user interaction rewards and the adjustment of their social interaction information can motivate users to participate in more interactions, especially when users see that their behavior receives real-time feedback and affects their social level, they will participate in social activities more actively; in addition, regularly adjusting the interaction level based on the value-added of rewards helps to enhance users' sense of belonging and achievement and improve their loyalty;

[0085] By calculating the reward value, the value of the user's interactive behavior in the virtual environment can be quantified; for example, in a social platform, the user's participation in discussions, posting content or liking can be converted into reward values ​​to help the system evaluate the user's activity, contribution or influence; the reward value is the basis for the system to design an incentive mechanism. By analyzing the user's reward value-added, the platform can design more targeted incentive strategies, such as customized rewards, virtual items or coupons, to stimulate users to participate more actively; the analysis of reward values ​​can help the platform identify which interactive behaviors can effectively improve user experience and satisfaction; by continuously optimizing the calculation method of reward values, the platform can provide services that better meet user needs, thereby enhancing user stickiness and loyalty; reward values ​​can also be converted into physical rewards to increase user interaction enthusiasm.

[0086] Example 2, a user interaction experience optimization system based on Metaverse, see Figure 1 As shown, including:

[0087] The interactive feature extraction module includes a data acquisition unit and a feature analysis unit; the data acquisition unit is used to acquire user virtual interaction information; the user virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information; the feature analysis unit is used to input the user virtual interaction information into the user interaction feature extraction model for feature extraction to obtain a user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modal feature recognition layer, an auditory modal feature recognition layer, a text modal feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and the multimodal feature analysis method is used in the user interaction feature extraction model to improve and analyze the convolutional neural network model;

[0088] The interactive reward experience optimization module includes a reward analysis unit and an information optimization unit; the reward analysis unit is used to perform feature recognition on the user virtual interaction information feature set based on the user virtual interaction information feature set and the user interaction reward analysis model to obtain the target user interaction reward value-added; the user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer, a reward value quantization function is deployed in the reward value analysis layer, the group optimization algorithm is improved and updated in the reward value analysis layer, and a reward value quantization function is constructed; the information optimization unit is used to update the user social interaction information based on the target user interaction reward value-added every preset time range.

[0089] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for optimizing user interaction experience based on Metaverse, characterized in that: The following steps are involved: Obtain user virtual interaction information; User virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information; The user virtual interaction information is input into the user interaction feature extraction model for feature extraction to obtain the user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modality feature recognition layer, an auditory modality feature recognition layer, a text modality feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and the multimodal feature analysis method is used in the user interaction feature extraction model to improve and analyze the convolutional neural network model; Based on the user virtual interaction information feature set and the user interaction reward analysis model, feature recognition is performed on the user virtual interaction information feature set to obtain the target user interaction reward increment; the user social interaction information is updated based on the target user interaction reward increment every preset time range; The user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer. A reward value quantization function is deployed in the reward value analysis layer. In the reward value analysis layer, the group optimization algorithm is improved and updated to construct a reward value quantization function.

2. A user interaction experience optimization method based on metaverse according to claim 1, characterized in that: The user interaction feature extraction model includes a data classification layer, a visual modality feature recognition layer, an auditory modality feature recognition layer, a text modality feature recognition layer, a social information feature recognition layer, and a multimodal feature coupling layer; The data classification layer is used to input the user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information in the user virtual interaction information into the corresponding feature recognition layer for analysis; Among them, the user visual interaction information is input into the visual modality feature recognition layer, the user auditory interaction information is input into the auditory modality feature recognition layer, the user text interaction information is input into the text modality feature recognition layer, and the user social interaction information is input into the social information feature recognition layer; The visual modality feature recognition layer is used to extract multiple features of user visual interaction information to obtain user visual interaction fusion features; The auditory modality feature recognition layer is used to extract Mel-frequency cepstral coefficients from the user auditory interaction information using distributed discrete cosine transform to obtain the user auditory interaction features; The text modality feature recognition layer is used to extract text semantic representation from user text interaction information and obtain user text interaction features; The social information feature recognition layer is used to extract features of user social interaction information and obtain user social interaction features; The multimodal feature coupling layer is used to couple the user visual interaction fusion features, user auditory interaction features, user text interaction features and user social interaction features using Bayesian expert product to obtain the user virtual interaction information feature set.

3. A user interaction experience optimization method based on metaverse according to claim 2, characterized in that: The specific steps of performing multi-feature extraction in the visual modality feature recognition layer include: Three feature extraction convolution blocks and a feature fusion module are constructed in the visual modality feature recognition layer; the first dense convolution block is used to extract the global features of the user visual interaction information to obtain the global features of the user visual interaction; the second dense convolution block is used to extract the emotional features of the user visual interaction information to obtain the emotional features of the user visual interaction; the third dense convolution block is used to extract the action features of the user visual interaction information to obtain the action features of the user visual interaction; The specific steps of constructing any one of the three feature extraction convolution blocks include: A convolutional layer is constructed using a single multi-box detector, and multiple convolutional layers are connected to construct a feature extraction convolutional block; the corresponding sub-features are output in each convolutional layer, and the output features of the previous convolutional layer are used as the input features of the next convolutional layer. After all sub-features are average-pooled in the last convolutional layer, the output features of the corresponding feature extraction convolutional block are obtained; In the feature fusion module, the user visual interaction global features, the user visual interaction emotional features and the user visual interaction action features are fused to obtain the user visual interaction fusion features.

4. The method for optimizing user interaction experience based on Metaverse according to claim 3, characterized in that: The specific steps of extracting text semantic representation from user text interaction information in the text modality feature recognition layer include: A pre-trained semantic analysis model is constructed, and a pre-trained sentiment recognition model is connected to the pre-trained semantic analysis model to form a text modal feature recognition layer; in the pre-trained semantic analysis model, semantic vector analysis is performed on user text interaction information to obtain a number of user text interaction vectors; in the pre-trained sentiment recognition model, sentiment recognition is performed on each user text interaction vector, and sentiment labels are added to obtain a number of user text interaction vectors with sentiment labels; all user text interaction vectors with sentiment labels are connected to obtain user text interaction features.

5. The method for optimizing user interaction experience based on Metaverse according to claim 4, characterized in that: The user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer, and a result output layer; The data preprocessing layer is used to preprocess the user virtual interaction information feature set to obtain the preprocessed virtual interaction information feature value T n , n=1,2,…,N; N is the total number of features in the user virtual interaction information feature set; The reward value analysis layer is used to calculate the reward value based on all preprocessed virtual interaction information feature values ​​T n Analyze with the reward value quantization function to obtain the target user interaction reward increment; The result output layer is used to output the target user interaction reward increment.

6. A method for optimizing user interaction experience based on metaverse according to claim 5, characterized in that: The specific steps to construct the reward value quantization function in the reward value regression analysis layer include: Collect feature recognition reward increment data sets; the feature recognition reward increment data sets contain several sets of feature value sets with target value annotations, and the target value corresponds to the verified reward increment; Construct K simulated reward value quantization function individuals H based on feature recognition reward increment dataset k ; Construct the initial individual position matrix; The initial individual position matrix is ​​used to represent each simulated reward value quantization function individual H k The initial position of; K simulated reward value quantization function individual H k Combine to obtain the reward value quantization function iterative population; set the maximum number of iterations; The specific steps of calculating the fitness of the individual of the simulated reward value quantization function are: using the individual of the simulated reward value quantization function to calculate the feature recognition reward increment data set to obtain the simulated feature increment; calculating the Euclidean distance between the simulated feature increment and the target value in the feature recognition reward increment data set to obtain the simulated calculation difference; and taking the inverse of the simulated calculation difference as the fitness of the individual of the simulated reward value quantization function; Set the position update formula F(t+1), F(t+1)=F(t)+ Q; where t represents the current iteration number, and F(t) represents the initial individual position matrix when the current iteration number is t; Q represents random update of individual positions; Based on the position update formula F(t+1), the initial individual position matrix is ​​updated to obtain a new initial individual position matrix; based on the new initial individual position matrix, the reward value quantization function iteration population is iterated to obtain a new reward value quantization function iteration population; When the maximum number of iterations is reached, the simulated reward value quantization function individual corresponding to the maximum output fitness is the optimal simulated reward value quantization function individual; the reward value quantization function in the reward value regression analysis layer is constructed based on the optimal simulated reward value quantization function individual.

7. A method for optimizing user interaction experience based on Metaverse according to claim 6, characterized in that: The specific steps of updating the user social interaction information based on the target user interaction reward increment every preset time range include: Calculate the total amount of all target user interaction reward appreciation within the preset time range to obtain the stage target user interaction reward appreciation; match the stage target user interaction reward appreciation in the user interaction implicit value matching library to obtain the corresponding updated target user interaction level; update the user social interaction information based on the updated target user interaction level to obtain new user social interaction information.

8. A user interaction experience optimization system based on Metaverse, characterized in that: The system applies a user interaction experience optimization method based on the metaverse as described in any one of claims 1 to 7, including: The interactive feature extraction module includes a data acquisition unit and a feature analysis unit; the data acquisition unit is used to acquire user virtual interaction information; the user virtual interaction information includes user visual interaction information, user auditory interaction information, user text interaction information and user social interaction information; the feature analysis unit is used to input the user virtual interaction information into the user interaction feature extraction model for feature extraction to obtain a user virtual interaction information feature set; the user interaction feature extraction model includes a data classification layer, a visual modal feature recognition layer, an auditory modal feature recognition layer, a text modal feature recognition layer, a social information feature recognition layer and a multimodal feature coupling layer, and the multimodal feature analysis method is used in the user interaction feature extraction model to improve and analyze the convolutional neural network model; The interactive reward experience optimization module includes a reward analysis unit and an information optimization unit; the reward analysis unit is used to perform feature recognition on the user virtual interaction information feature set based on the user virtual interaction information feature set and the user interaction reward analysis model to obtain the target user interaction reward value-added; the user interaction reward analysis model includes a data preprocessing layer, a reward value analysis layer and a result output layer, a reward value quantization function is deployed in the reward value analysis layer, the group optimization algorithm is improved and updated in the reward value analysis layer, and a reward value quantization function is constructed; the information optimization unit is used to update the user social interaction information based on the target user interaction reward value-added every preset time range.

Images