Virtual reality natural language interaction control system based on big data

Abstract

The invention relates to the field of language control systems, and discloses a virtual reality natural language interaction control system based on big data, and the system comprises a data acquisition unit which collects the interaction original data of a virtual reality scene according to a sensor array; performing natural language semantic analysis on the interaction original data to obtain user intention semantic feature data; performing virtual scene correlation analysis according to the user intention semantic feature data to obtain scene correlation data; a data prediction unit; through natural language semantic analysis in the data acquisition unit, intention feature data of the user can be extracted, so that the system can more accurately understand the requirements of the user in a virtual reality environment, the naturalness and fluency of interaction are improved, and in addition, by combining scene correlation analysis and user behavior pattern mining, the interaction efficiency is improved. The interactive intention of the user can be dynamically predicted, so that the user demand is responded in advance, and the immersive experience is enhanced.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of language control systems, in particular to a virtual reality natural language interaction control system based on big data. BACKGROUND

[0002] The virtual reality natural language interaction of big data combines virtual reality and natural language processing technology, and uses big data analysis to realize more intelligent and personalized virtual environment interaction. Through the processing capacity of big data, the speech, action and emotion of a user can be comprehensively analyzed, and a more accurate and smooth virtual reality experience is provided for the user.

[0003] The existing device has the following disadvantages: the traditional virtual reality system often relies on preset scripts and fixed response rules, lacks accurate understanding of user intentions and emotional states, and therefore the interaction process can be relatively rigid, cannot effectively predict user needs, results in poor immersion, lacks a dynamic adjustment mechanism in the adaptation degree evaluation of user historical data and physiological signals, cannot be optimized according to the unique needs and emotional states of each user, and therefore reduces personalized experience and emotional resonance, usually relies on a single computing node to process data, can cause delay or instability in complex interaction and big data scenarios, cannot quickly respond to user behavior, affects system smoothness and user experience, and relies on manual configuration and intervention, and lacks an adaptive mechanism.

[0004] Therefore, the virtual reality natural language interaction control system based on big data is proposed to solve the above problems. SUMMARY

[0005] The application aims to provide a virtual reality natural language interaction control system based on big data to solve the problem that the traditional virtual reality system often relies on preset scripts and fixed response rules, lacks accurate understanding of user intentions and emotional states, and therefore the interaction process can be relatively rigid and cannot effectively predict user needs.

[0006] To achieve the above-mentioned purpose, the application provides the following technical scheme: a virtual reality natural language interaction control system based on big data, comprising: A data acquisition unit acquires interaction raw data of a virtual reality scene according to a sensor array, performs natural language semantic analysis on the interaction raw data to obtain user intention semantic feature data, and performs virtual scene correlation analysis according to the user intention semantic feature data to obtain scene correlation data. The data prediction unit mines user behavior patterns according to the scene correlation data to obtain user behavior pattern data; conducts dynamic prediction of an interactive intention in combination with the user behavior pattern data and the user intention semantic feature data to obtain interactive intention prediction data; generates a virtual reality scene response strategy according to the interactive intention prediction data to obtain scene response strategy data; The data correction unit evaluates the adaptation degree of the scene response strategy data according to the user historical interaction data to obtain adaptation degree data; and corrects the adaptation degree data in combination with the physiological signal data to obtain the adaptation degree data after emotional correction; The model optimization unit trains an interactive control model through an algorithm on the adaptation degree data after emotional correction to obtain a virtual reality interactive control model; deploys the virtual reality interactive control model to an edge computing node and conducts dynamic optimization of the model to obtain an optimized interactive control model; According to the optimized interactive control model, the virtual reality scene is adjusted in parameters to obtain scene control instruction data; and the scene control instruction data is synchronized to a cloud platform to execute natural language interactive control of the virtual reality environment.

[0007] According to the sensor array, interactive original data of the virtual reality scene are collected; natural language semantic analysis is conducted on the interactive original data to obtain user intention semantic feature data, including: The multi-modal interactive original data stream is preprocessed, wherein the preprocessing includes noise filtering, timestamp alignment and data format standardization to obtain preprocessed synchronous data stream; According to the preprocessed synchronous data stream, feature extraction is conducted to obtain a feature vector set; according to the current virtual reality scene type, a weight coefficient of a modal feature vector is dynamically allocated to obtain a weighted multi-modal feature set; The weighted multi-modal feature set is input into a pre-trained cross-modal semantic analysis model and feature correlation analysis is conducted to obtain preliminary user intention semantic features; the preliminary user intention semantic features are corrected in individual preference through a user historical interaction behavior database to obtain corrected user intention semantic features; Based on the corrected user intention semantic features, a multi-modal intention Figure One The consistency verification model identifies and integrates conflict information between modes to obtain final user intention semantic feature data.

[0008] According to the user intention semantic feature data, virtual scene correlation analysis is conducted to obtain scene correlation data, including: A virtual scene dynamic element library is constructed, and an interactive object, an environment parameter and an event trigger condition in a virtual reality scene are extracted, and scene element graph is obtained by combining user historical interactive scene preference data; the data source of the scene element graph is a scene type determined by a weighted multi-modal feature set; The user intention semantic feature data is mapped to the scene element graph, and scene objects and events with strong association with the intention keywords are identified by semantic similarity calculation; a attention mechanism is introduced to strengthen the weight of the scene elements with high matching degree, and an initial scene association candidate set is generated.

[0009] Among them, the initial scene association candidate set is dynamically adjusted based on user behavior feedback, and low correlation elements are removed and the weight of high correlation elements is enhanced to form a dynamic scene association weight matrix; the data source of the user real-time behavior feedback is a multi-modal interaction original data stream collected by a sensor array; The dynamic scene association weight matrix is verified for consistency by a multi-modal association verification model to obtain scene association data.

[0010] Among them, according to the scene association data, user behavior pattern data is obtained, including: A user behavior sequence library is constructed, and the behavior rules of the user are recorded based on the scene association data to obtain a basic behavior sequence library; According to the dynamic scene association weight matrix, the user's high-frequency operation mode is extracted, and the intention keywords in the user intention semantic feature data are combined to obtain a candidate behavior pattern cluster by an unsupervised clustering algorithm; A time series attention mechanism is introduced to analyze the time dependence of the candidate behavior pattern cluster to obtain a structured behavior pattern template; The structured behavior pattern template is dynamically verified by a reinforcement learning model, simulating the user's potential operation path in different scenes, and adjusting the template weight by comparing the user behavior feedback data to obtain user behavior pattern data.

[0011] Among them, the user behavior pattern data and the user intention semantic feature data are combined to dynamically predict the interactive intention, and the interactive intention prediction data is obtained, including: A behavior intention fusion feature space is constructed, and the user behavior pattern data and the user intention semantic feature data are associated by the behavior intention fusion feature space to obtain an initial intention association graph; According to the initial intention association graph, a scene adaptation degree is introduced to filter incompatible behavior patterns, a behavior pattern execution confidence model is constructed by combining user real-time behavior feedback data, and the behavior template weight is dynamically adjusted to obtain a weighted intention candidate set; inputting the set of weighted intention candidates into a spatio-temporal graph neural network, combining spatial positions and event trigger conditions in the scene correlation data, predicting an evolution path of the user intention in a time dimension to obtain a spatio-temporal intention trajectory; comparing logical consistencies of the voice instruction, the behavior trajectory and the scene event trigger result through a modal causal verification engine, correcting prediction deviations caused by semantic ambiguity and behavior inertia to obtain interaction intention prediction data.

[0012] According to the interaction intention prediction data, a virtual reality scene response strategy is generated to obtain scene response strategy data, including: parsing an intention target, an operation sequence and scene dependency in the interaction intention prediction data, combining a scene element atlas in the scene correlation data to obtain an initial scene response strategy candidate set; introducing an operation path in the user behavior pattern data, dynamically adjusting priority of the strategy candidate set in combination with real-time scene data, and screening a strategy template through a reinforcement learning model; fusing the strategy template and the intention evolution path to construct a spatio-temporal continuous scene response strategy tree, labeling a trigger condition, an execution action and an expected effect of each node to obtain a response strategy chain; comparing matching degrees of the response strategy chain and the interaction intention prediction data through a modal causal verification engine to obtain scene response strategy data.

[0013] According to the user historical interaction data, the scene response strategy data is adaptively evaluated to obtain adaptability data; the adaptability data is modified in terms of emotional state in combination with physiological signal data to obtain emotion-modified adaptability data, including: extracting strategy execution records and preference labels in the user historical interaction data to construct a historical strategy adaptability benchmark model; inputting the scene response strategy data into the model to calculate a behavior pattern matching degree to obtain an initial adaptability score; loading physiological signal data, mapping the physiological signal data into emotional labels through an emotional recognition model; dynamically adjusting adaptability score weights in combination with the emotional labels to obtain emotion-modified adaptability data; constructing an emotional strategy correlation atlas; introducing a reinforcement learning model to simulate emotional feedback of the user on the emotional strategy to dynamically optimize adaptability score strategies; comparing causal relationships between the emotion-modified adaptability data and real-time scene response effects through a modal verification engine to obtain emotion-modified adaptability data.

[0014] Specifically, the interaction control model is trained on the emotion-corrected fit data using an algorithm to obtain a virtual reality interaction control model; the virtual reality interaction control model is deployed to edge computing nodes and dynamically optimized to obtain an optimized interaction control model, including: Construct a structured training dataset for modal features; extract key feature vectors from the structured training dataset to obtain model input samples; We design a meta-learning and reinforcement learning hybrid training framework, optimize the interactive control policy through the policy gradient algorithm, and introduce a transfer learning mechanism to initialize model parameters using historical user data. The trained interactive control model is deployed to edge computing nodes, and the consumption of computing resources is reduced through model quantization and pruning techniques. An edge-cloud collaborative architecture is built, user interaction data is uploaded to the cloud for global model updates, and optimization parameters are synchronously distributed to edge nodes. Based on the adaptation data collected from edge nodes and user feedback, the model hyperparameters are dynamically adjusted, an adversarial training mechanism is introduced to enhance the model robustness, and the model decision boundary is optimized by simulating abnormal interaction scenarios to obtain an optimized interaction control model.

[0015] The initial intent association graph construction formula is as follows: in, Indicates the first The user and the The correlation or similarity between items Indicates user The weight matrix, Indicates the first Feature vectors of each user Represents items The weight matrix, Indicates the first The feature vector of an item.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention can extract user intent feature data through natural language semantic parsing in the data acquisition unit. This enables the system to more accurately understand the user's needs in the virtual reality environment, improve the naturalness and fluency of the interaction. Moreover, by combining scene correlation analysis and user behavior pattern mining, it can dynamically predict the user's interaction intent, thereby responding to user needs in advance and enhancing the immersive experience. 2、The system can adapt to the user's historical interaction data and physiological signal data for adaptability evaluation and emotional correction, ensure that the virtual reality scene response strategy meets the personalized needs of each user, and the emotional state correction further improves the accuracy of interaction and emotional resonance, so that the virtual environment and the emotional needs of the user are more matched, and the affinity and participation of the user experience are improved; 3、The interactive control model training and dynamic optimization by the model optimization unit can be performed in real time on the edge computing node, ensuring fast response and low delay of the system, which is very important for scene control in virtual reality with high real-time requirements, especially in complex dynamic interaction and big data processing environment, which can effectively improve the reaction speed and flexibility of the system; 4、Based on the optimized interactive control model, the system can automatically adjust the parameters of the virtual reality scene to ensure high matching between the environment and the user's needs, which not only improves the adaptability of virtual reality experience, but also reduces the need for human intervention, making the interaction process smoother and more intelligent; 5、The whole system can timely transmit and execute scene control instruction data through the synchronization of the cloud platform, realizing scene control based on big data driving, which not only improves the intelligent level of virtual reality environment, but also enhances the adaptability of the system in different scenes, so that it can automatically optimize the interactive experience in different use situations; 6、The system can perceive the emotional changes of the user and adjust the interactive response of the virtual scene in time by combining physiological signal data for emotional correction, which is very important for emotional interaction in virtual reality. Such emotional regulation makes the user obtain more rich and real emotional feedback in the experience, improving the emotional participation of virtual reality application. The system cooperates with the edge computing node through the cloud platform, so that the model can be continuously updated and optimized. The design of such architecture enhances the scalability and adaptability of the system, so that it can better meet the needs of virtual reality application in different scenes and complexity. Through the combination of edge computing and cloud platform, the optimal allocation and efficient operation of resources can be realized in a distributed environment, reducing the operation and maintenance cost and the consumption of computing resources. At the same time, dynamic optimization and model adaptation also reduce the need for human intervention, reducing the complexity of system optimization and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the system architecture structure of the whole system in an embodiment of the application.

[0018] In the figure: 1, data acquisition unit; 2, data prediction unit; 3, data correction unit; 4, model optimization unit. DETAILED DESCRIPTION

[0019] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application. Embodiment one

[0020] Please refer to Figure 1 The present application provides a technical solution: a virtual reality natural language interaction control system based on big data, comprising: A data acquisition unit 1 acquires interaction raw data of a virtual reality scene according to a sensor array; performs natural language semantic analysis on the interaction raw data to obtain user intention semantic feature data; and performs virtual scene correlation analysis according to the user intention semantic feature data to obtain scene correlation data. A data prediction unit 2 mines user behavior pattern data according to the scene correlation data; performs dynamic prediction of interaction intention in combination with the user behavior pattern data and the user intention semantic feature data to obtain interaction intention prediction data; generates virtual reality scene response strategies according to the interaction intention prediction data to obtain scene response strategy data; A data correction unit 3 evaluates the adaptation degree of the scene response strategy data according to user historical interaction data to obtain adaptation degree data; and corrects the adaptation degree data in combination with physiological signal data to obtain emotion-corrected adaptation degree data. A model optimization unit 4 trains an interaction control model for the emotion-corrected adaptation degree data through an algorithm to obtain a virtual reality interaction control model; deploys the virtual reality interaction control model to an edge computing node and performs dynamic optimization of the model to obtain an optimized interaction control model. According to the optimized interaction control model, parameters of the virtual reality scene are adjusted to obtain scene control instruction data; and the scene control instruction data is synchronized to a cloud platform to perform natural language interaction control of the virtual reality environment.

[0021] It is necessary to explain that in operation, the data acquisition unit: through the sensor array, real-time collection of interactive raw data in the virtual reality scene. These data include user's actions, voice, position and other information, the purpose is to obtain comprehensive user behavior data. These raw data are parsed through natural language processing technology to obtain user's intention semantic feature data. This process is the key to understanding user needs, ensuring that the system can make appropriate responses according to user input. According to the user's intention semantic feature data, the system analyzes the relevance between the current virtual scene and the user's demand. This helps the system understand the state of the current scene and the user's expectations, so as to make a more demand-oriented response. The system conducts pattern mining on user behavior according to scene relevance data, analyzes user's behavior habits and preferences in similar situations. This provides data support for predicting future interaction behavior. Combined with user's historical behavior and intention semantic features, the system can dynamically predict the user's interaction intention. This enables the system to respond in advance, improving the smoothness of interaction. According to the predicted interaction intention, the system generates a response strategy for the virtual reality scene. This is a specific rule for the system to respond to user behavior in the virtual scene. Through the user's historical interaction data, the system evaluates the adaptation degree of the response strategy, and combines the user's physiological signals (such as heart rate, emotion, etc.) to make emotional correction. This ensures that the response not only meets the functional requirements, but also takes into account the emotional needs of the user. The system trains the corrected adaptation degree data through algorithm to generate an interaction control model. The model can adapt to user behavior and needs in real time. Through real-time calculation and optimization to reduce delay and improve response speed. The role of edge computing is to ensure that data processing and scene adjustment can be quickly executed on the user's device, thereby avoiding the delay caused by cloud computing. The optimized interaction control model adjusts the parameters of the virtual scene according to the predicted needs. This process involves changing elements in the virtual scene to better adapt to the user's intention. All scene control instructions are executed synchronously through the cloud platform to ensure that changes in the virtual reality environment can be reflected in time, realizing seamless interaction. Through natural language processing and behavior pattern prediction, the system can more accurately identify user needs and improve the naturalness of interaction. The system can adjust the response strategy according to each user's historical behavior and emotional state, so that each user's virtual reality experience can best meet their individual needs. Through real-time optimization model and edge computing, the system can provide low-latency, high-response virtual reality experience. The system can automatically adjust the scene according to the user's emotional changes and needs to realize more personalized interaction. Embodiment two

[0022] In an embodiment, interactive raw data of a virtual reality scene is collected according to a sensor array; natural language semantic analysis is performed on the interactive raw data to obtain user intention semantic feature data, including: Preprocessing the multimodal interaction raw data stream, wherein the preprocessing includes noise filtering, timestamp alignment, and data format standardization, to obtain a preprocessed synchronized data stream; Performing feature extraction according to the preprocessed synchronized data stream to obtain a feature vector set; dynamically assigning weight coefficients of modal feature vectors according to the current virtual reality scene type to obtain a weighted multimodal feature set; Inputting the weighted multimodal feature set into a pre-trained cross-modal semantic parsing model and performing feature correlation analysis to obtain preliminary user intent semantic features; performing personalized preference correction on the preliminary user intent semantic features through a user historical interaction behavior database to obtain corrected user intent semantic features; Based on the corrected user intent semantic features, constructing a multimodal intent Figure One The consistency verification model identifies and integrates conflict information between modalities to obtain final user intent semantic feature data.

[0023] In this way, the user's interaction data is collected by a sensor array, noise filtering, timestamp alignment, and format standardization are performed to obtain a synchronized data stream. Then, the system extracts feature vectors and dynamically adjusts the weights according to the virtual reality scene to generate a weighted multimodal feature set. This feature set is input into a cross-modal semantic parsing model for feature correlation analysis to obtain preliminary user intent semantic features. Then, the user intent features are corrected through user historical behavior data to finally obtain personalized user intent semantic features. Finally, a multimodal intent Figure One The consistency verification model integrates conflict information to ensure that the system can accurately understand and respond to user needs, optimizing the virtual reality interactive experience.

[0024] In an embodiment, virtual scene correlation analysis is performed according to the user intent semantic feature data to obtain scene correlation data, including: A virtual scene dynamic element library is constructed to extract interactive objects, environmental parameters, and event trigger conditions in the virtual reality scene, and combined with user historical interaction scene preference data to obtain a scene element graph; the data source of the scene element graph is the scene type determined by the weighted multimodal feature set; The user intent semantic feature data is mapped to the scene element graph to identify scene objects and events with strong correlation to the intent keywords through semantic similarity calculation; an attention mechanism is introduced to weight reinforce scene elements with high matching degree to generate an initial scene correlation candidate set; Based on user behavior feedback, the initial scene correlation candidate set is dynamically adjusted to eliminate low correlation elements and enhance the weights of high correlation elements to form a dynamic scene correlation weight matrix; the data source of the user real-time behavior feedback is the multimodal interaction raw data stream collected by the sensor array; The dynamic scene correlation weight matrix is verified for consistency by a multi-modal correlation verification model to obtain scene correlation data.

[0025] In this way, a virtual scene dynamic element library is constructed, interactive objects, environmental parameters and event conditions in the scene are extracted, and scene element atlas is generated in combination with user historical interaction data. According to user intention feature data, the user intention feature data is mapped to the scene element atlas, and scene objects and events related to the user intention are identified through semantic similarity calculation. An attention mechanism is introduced to strengthen high matching degree elements, and an initial scene correlation candidate set is generated. Then, the candidate set is dynamically adjusted through user behavior feedback data, and the correlation weight matrix is optimized. Finally, a multi-modal correlation verification model is used to verify and optimize the weight matrix, ensuring the accuracy of the scene correlation data. This process helps the system to respond more accurately to user needs, improving the immersion and precision of virtual reality interaction. Embodiment Three

[0026] In an embodiment, user behavior pattern data is obtained according to scene correlation data, including: A user behavior sequence library is constructed to record user behavior patterns based on scene correlation data to obtain a basic behavior sequence library; According to the dynamic scene correlation weight matrix, user high-frequency operation patterns are extracted, combined with intention keywords in user intention semantic feature data, and through unsupervised clustering algorithm, candidate behavior pattern clusters are obtained; A time series attention mechanism is introduced to analyze the time dependence of the candidate behavior pattern clusters to obtain a structured behavior pattern template; A reinforcement learning model is used to dynamically verify the structured behavior pattern template, simulate the user's potential operation path in different scenes, and adjust the template weight by comparing user behavior feedback data to obtain user behavior pattern data.

[0027] In this way, a user behavior sequence library is constructed to record user behavior patterns in virtual scenes. Then, based on the dynamic scene correlation weight matrix, user high-frequency operation patterns are extracted, combined with intention keywords, and through unsupervised clustering algorithm, candidate behavior pattern clusters are generated. Then, the time series attention mechanism is used to analyze the time dependence of these clusters to obtain a structured behavior pattern template. Finally, the reinforcement learning model is used to dynamically verify the template, simulate the user's potential operation path in different scenes, and adjust the template weight by comparing with user behavior feedback data, so as to obtain more accurate user behavior pattern data. This process helps to improve the system's prediction ability, making virtual reality experience more personalized and smooth.

[0028] In an embodiment, the interaction intention dynamic prediction is performed in combination with the user behavior pattern data and the user intention semantic feature data to obtain interaction intention prediction data, including: A behavior intention fusion feature space is constructed, and the user behavior pattern data and the user intention semantic feature data are associated through the behavior intention fusion feature space to obtain an initial intention association graph; According to the initial intention association graph, a scene adaptation degree is introduced to filter incompatible behavior patterns, a behavior pattern execution confidence model is constructed in combination with real-time user behavior feedback data, and a weighted intention candidate set is obtained by dynamically adjusting the behavior template weight; The weighted intention candidate set is input into a spatio-temporal graph neural network, the spatial position and event trigger condition in the scene association data are combined, the evolution path of the user intention in the time dimension is predicted, and a spatio-temporal intention trajectory is obtained; Through a modal causal verification engine, the logical consistency of the voice instruction, the behavior trajectory, and the scene event trigger result is compared, the prediction deviation caused by semantic ambiguity and behavior inertia is corrected, and the interaction intention prediction data is obtained.

[0029] In this way, the behavior intention fusion feature space is constructed, the user behavior pattern data and the intention feature data are associated, and the initial intention association graph is generated. Then, based on the graph, the scene adaptation degree is introduced to filter incompatible behavior patterns, and the behavior pattern execution confidence model is constructed in combination with real-time user behavior feedback. The weight is dynamically adjusted to form a weighted intention candidate set. Subsequently, the candidate set is input into a spatio-temporal graph neural network, the spatial position and event trigger condition are combined, the evolution path of the user intention in the time dimension is predicted, and a spatio-temporal intention trajectory is obtained. Finally, through a modal causal verification engine, the voice instruction, the behavior trajectory, and the scene event are compared, the prediction deviation is corrected, and the accuracy is ensured. This process helps to accurately predict user intentions and provides more personalized and smooth virtual reality interaction experiences.

[0030] In an embodiment, a virtual reality scene response strategy is generated according to the interaction intention prediction data to obtain scene response strategy data, including: The intention target, operation sequence, and scene dependency in the interaction intention prediction data are parsed, and the scene element graph in the scene association data is combined to obtain an initial scene response strategy candidate set; The operation path in the user behavior pattern data is introduced, the priority of the strategy candidate set is dynamically adjusted in combination with real-time scene data, and a strategy template is selected through a reinforcement learning model; The strategy template and the intention evolution path are fused to construct a spatio-temporal continuous scene response strategy tree, and the trigger condition, execution action, and expected effect of each node are labeled to obtain a response strategy chain; Through a modal causal verification engine, the matching degree of the response strategy chain and the interaction intention prediction data is compared to obtain the scene response strategy data.

[0031] In this way, the interaction intention prediction data is analyzed to extract the intention target, operation sequence and scene dependency, and an initial scene response strategy candidate set is generated in combination with a scene element atlas. Then, based on the operation path in the user behavior pattern data, the strategy priority is dynamically adjusted in combination with real-time scene data, and a suitable strategy template is selected through a reinforcement learning model. Next, the strategy template is fused with the intention evolution path to construct a time and space continuous scene response strategy tree, and the trigger condition, execution action and expected effect of each node are labeled to form a response strategy chain. Finally, the matching degree of the response strategy chain and the interaction intention prediction data is compared through a modal causal verification engine to ensure the accuracy of the response strategy. This method helps to improve the adaptability and accuracy of the virtual reality system and provides a more intelligent user experience.

[0032] In an embodiment, the scene response strategy data is evaluated for adaptation degree based on user historical interaction data to obtain adaptation degree data; the adaptation degree data is corrected for emotional state in combination with physiological signal data to obtain emotion-corrected adaptation degree data, including: The strategy execution records and preference labels in the user historical interaction data are extracted to construct a historical strategy adaptation degree benchmark model; the scene response strategy data is input into the model to calculate the behavior pattern matching degree to obtain an initial adaptation degree score; The physiological signal data is loaded, and the physiological signal data is mapped to emotional labels through an emotion recognition model; the adaptation degree score weight is dynamically adjusted in combination with the emotional labels to obtain emotion-corrected adaptation degree data; An emotional strategy association atlas is constructed; a reinforcement learning model is introduced to simulate the emotional feedback of the user to the emotional strategy to dynamically optimize the adaptation degree score strategy; The causal relationship between the emotion-corrected adaptation degree data and the real-time scene response effect is compared through a modal verification engine to obtain emotion-corrected adaptation degree data.

[0033] In this way, the strategy execution records and preference labels are extracted from the user historical interaction data to construct a benchmark model and calculate an initial adaptation degree score. Then, the physiological signal data is loaded, and the physiological signal data is converted to emotional labels through an emotion recognition model, and the adaptation degree score weight is dynamically adjusted in combination with these emotional labels to obtain emotion-corrected adaptation degree data. Next, an emotional strategy association atlas is constructed, and a reinforcement learning model is used to simulate the feedback of the user to the emotional strategy to optimize the adaptation degree score strategy. Finally, the causal relationship between the emotion-corrected adaptation degree data and the scene response effect is compared through a modal verification engine to ensure the accurate adjustment of the system to the emotional state of the user. This method helps to improve the response capability of the virtual reality system in the emotional aspect and enhances the user immersion.

[0034] In one embodiment, an interaction control model is trained on emotion-corrected fit data using an algorithm to obtain a virtual reality interaction control model; the virtual reality interaction control model is then deployed to an edge computing node and dynamically optimized to obtain an optimized interaction control model, including: Construct a structured training dataset for modal features; extract key feature vectors from the structured training dataset to obtain model input samples; We design a meta-learning and reinforcement learning hybrid training framework, optimize the interactive control policy through the policy gradient algorithm, and introduce a transfer learning mechanism to initialize model parameters using historical user data. The trained interactive control model is deployed to edge computing nodes, and the consumption of computing resources is reduced through model quantization and pruning techniques. An edge-cloud collaborative architecture is built, user interaction data is uploaded to the cloud for global model updates, and optimization parameters are synchronously distributed to edge nodes. Based on the adaptation data collected from edge nodes and user feedback, the model hyperparameters are dynamically adjusted, an adversarial training mechanism is introduced to enhance the model robustness, and the model decision boundary is optimized by simulating abnormal interaction scenarios to obtain an optimized interaction control model.

[0035] This design involves constructing a structured training dataset of modal features and extracting key feature vectors as input samples for the model. Next, a meta-learning and reinforcement learning hybrid training framework is designed. The interaction control strategy is optimized through a policy gradient algorithm, and a transfer learning mechanism is introduced to initialize model parameters using historical user data. Subsequently, the trained model is deployed to edge computing nodes. Model quantization and pruning techniques are used to reduce computational resource consumption, and an edge-cloud collaborative architecture is constructed. User interaction data is uploaded to the cloud for global updates, and optimized parameters are synchronized to the edge nodes. Finally, based on the edge node adaptation data and user feedback, the model hyperparameters are dynamically adjusted. Adversarial training is used to enhance model robustness, and abnormal interaction scenarios are simulated to optimize the decision boundary, resulting in an optimized interaction control model. This process helps improve the real-time responsiveness of the virtual reality system and the stability of the user experience.

[0036] In one embodiment, the initial intent association graph construction formula is as follows: in, Indicates the first The user and the The correlation or similarity between items Indicates user The weight matrix, Indicates the first Feature vectors of each user Represents items The weight matrix, represents a feature vector of the

[0037] The embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.​

Claims

1. A virtual reality natural language interaction control system based on big data, characterized in that, include: The data acquisition unit (1) collects the raw interactive data of the virtual reality scene based on the sensor array; Natural language semantic parsing is performed on the original interactive data to obtain user intent semantic feature data; Virtual scene correlation analysis is performed based on user intent semantic feature data to obtain scene correlation data; The data prediction unit (2) performs user behavior pattern mining based on the scene correlation data to obtain user behavior pattern data; combines user behavior pattern data with user intent semantic feature data to perform dynamic prediction of interaction intent to obtain interaction intent prediction data; and generates virtual reality scene response strategy based on interaction intent prediction data to obtain scene response strategy data. The data correction unit (3) evaluates the adaptability of the scenario response strategy data based on the user's historical interaction data to obtain adaptability data; The fitness data is then modified by combining physiological signal data to obtain emotionally modified fitness data. The model optimization unit (4) trains the interaction control model on the emotion-corrected adaptation data using an algorithm to obtain a virtual reality interaction control model; the virtual reality interaction control model is deployed to an edge computing node and dynamically optimized to obtain an optimized interaction control model. The parameters of the virtual reality scene are adjusted according to the optimized interactive control model to obtain scene control command data; the scene control command data is synchronized to the cloud platform to execute natural language interactive control of the virtual reality environment.

2. The virtual reality natural language interaction control system based on big data according to claim 1, characterized in that, Raw interactive data of virtual reality scenes are collected based on a sensor array; Natural language semantic parsing is performed on the raw interactive data to obtain user intent semantic feature data, including: The raw multimodal interaction data stream is preprocessed, including noise filtering, timestamp alignment, and data format standardization, to obtain a preprocessed synchronous data stream. Feature extraction is performed on the preprocessed synchronous data stream to obtain a set of feature vectors; weight coefficients of the modal feature vectors are dynamically assigned according to the current virtual reality scene type to obtain a weighted multimodal feature set; The weighted multimodal feature set is input into a pre-trained cross-modal semantic parsing model and feature association analysis is performed to obtain preliminary user intent semantic features. The preliminary user intent semantic features are then personalized and corrected using a user history interaction behavior database to obtain the corrected user intent semantic features. Based on the corrected user intent semantic features, a multimodal intent consistency verification model is constructed to identify and integrate conflict information between modalities, thereby obtaining the final user intent semantic feature data.

3. The virtual reality natural language interaction control system based on big data according to claim 2, characterized in that, Virtual scene correlation analysis is performed based on user intent semantic feature data to obtain scene correlation data, including: A dynamic element library for virtual scenes is constructed, and interactive objects, environmental parameters, and event triggering conditions in virtual reality scenes are extracted. Combined with user historical interaction scene preference data, a scene element map is obtained. The data source of the scene element map is the scene type determined by a weighted multimodal feature set. User intent semantic feature data is mapped to scene element graph, and scene objects and events strongly associated with intent keywords are identified through semantic similarity calculation; an attention mechanism is introduced to strengthen the weight of scene elements with high matching degree, and an initial scene relevance candidate set is generated.

4. The virtual reality natural language interaction control system based on big data according to claim 3, characterized in that, Virtual scene correlation analysis is performed based on user intent semantic feature data to obtain scene correlation data, which also includes: The initial scene relevance candidate set is dynamically adjusted based on user behavior feedback, low-relevance elements are removed and the weights of high-relevance elements are enhanced to form a dynamic scene relevance weight matrix; the data source of the real-time user behavior feedback is the multimodal interaction raw data stream collected by the sensor array. The consistency of the dynamic scene correlation weight matrix is ​​verified by using a multimodal correlation verification model to obtain scene correlation data.

5. The virtual reality natural language interaction control system based on big data according to claim 4, characterized in that, Based on the aforementioned scenario-related data, user behavior pattern mining is performed to obtain user behavior pattern data, including: Construct a user behavior sequence library by recording user behavior patterns based on the scenario-related data to obtain a basic behavior sequence library; Based on the dynamic scene correlation weight matrix, high-frequency user operation patterns are extracted. Combined with intent keywords in user intent semantic feature data, an unsupervised clustering algorithm is used to obtain candidate behavior pattern clusters. A time-series attention mechanism is introduced to perform time-dependency analysis on the candidate behavior pattern clusters to obtain structured behavior pattern templates; The structured behavior pattern template is dynamically validated by a reinforcement learning model to simulate the potential operation paths of users in different scenarios. The template weights are adjusted by comparing user behavior feedback data to obtain user behavior pattern data.

6. The virtual reality natural language interaction control system based on big data according to claim 5, characterized in that, By combining user behavior pattern data and user intent semantic feature data, dynamic prediction of interaction intent is performed to obtain interaction intent prediction data, including: Construct a behavioral intent fusion feature space, and associate user behavior pattern data with user intent semantic feature data through the behavioral intent fusion feature space to obtain an initial intent association map; Based on the initial intent association graph, scene adaptability is introduced to filter incompatible behavior patterns. Combined with real-time user behavior feedback data, a behavior pattern execution confidence model is constructed, and the weight of behavior templates is dynamically adjusted to obtain a weighted intent candidate set. The weighted intent candidate set is input into the spatiotemporal graph neural network, and combined with the spatial location and event triggering conditions in the scene correlation data, the evolution path of user intent in the time dimension is predicted to obtain the spatiotemporal intent trajectory. By using a modal causal verification engine, the logical consistency of voice commands, behavioral trajectories, and scene event triggering results is compared to correct prediction biases caused by semantic ambiguity and behavioral inertia, so as to obtain interactive intent prediction data.

7. The virtual reality natural language interaction control system based on big data according to claim 6, characterized in that, Virtual reality scene response strategies are generated based on interaction intent prediction data, resulting in scene response strategy data, including: The intention target, operation sequence and scene dependency in the interaction intention prediction data are analyzed and combined with the scene element map in the scene correlation data to obtain the initial scene response strategy candidate set; By incorporating operation paths from user behavior pattern data and dynamically adjusting the priority of the policy candidate set in conjunction with real-time scenario data, a reinforcement learning model is used to filter policy templates. By integrating the aforementioned strategy template and intent evolution path, a spatiotemporally continuous scene response strategy tree is constructed, and the triggering conditions, execution actions, and expected effects of each node are labeled to obtain the response strategy chain. The scene response strategy data is obtained by comparing the matching degree between the response strategy chain and the interaction intent prediction data through a modal causal verification engine.

8. The virtual reality natural language interaction control system based on big data according to claim 7, characterized in that, The adaptation of the scenario response strategy data is evaluated based on the user's historical interaction data to obtain adaptation data. The fitness data is then modified by combining physiological signal data to obtain emotionally modified fitness data, including: Extract strategy execution records and preference tags from user historical interaction data to construct a historical strategy fit benchmark model; input scenario response strategy data into the model and calculate the behavior pattern matching degree to obtain an initial fit score; Physiological signal data is loaded and mapped to emotion labels through an emotion recognition model; the fit score weights are dynamically adjusted in combination with the emotion labels to obtain emotion-corrected fit data. Construct an emotion strategy association graph; introduce a reinforcement learning model to simulate users' emotional feedback to emotion strategies and dynamically optimize the fit scoring strategy; The causal relationship between the emotion-corrected fit data and the real-time scene response effect is compared using a modal verification engine to obtain the emotion-corrected fit data.

9. The virtual reality natural language interaction control system based on big data according to claim 8, characterized in that, The interaction control model is trained on the emotion-corrected fit data using an algorithm to obtain a virtual reality interaction control model; the virtual reality interaction control model is then deployed to edge computing nodes and dynamically optimized to obtain an optimized interaction control model, including: Construct a structured training dataset for modal features; extract key feature vectors from the structured training dataset to obtain model input samples; We design a meta-learning and reinforcement learning hybrid training framework, optimize the interactive control policy through the policy gradient algorithm, and introduce a transfer learning mechanism to initialize model parameters using historical user data. The trained interactive control model is deployed to edge computing nodes, and the consumption of computing resources is reduced through model quantization and pruning techniques. An edge-cloud collaborative architecture is built, user interaction data is uploaded to the cloud for global model updates, and optimization parameters are synchronously distributed to edge nodes. Based on the adaptation data collected from edge nodes and user feedback, the model hyperparameters are dynamically adjusted, an adversarial training mechanism is introduced to enhance the model robustness, and the model decision boundary is optimized by simulating abnormal interaction scenarios to obtain an optimized interaction control model.

10. The virtual reality natural language interaction control system based on big data according to claim 9, characterized in that: The formula for constructing the initial intent association graph is as follows: in, Indicates the first The user and the The correlation or similarity between items Indicates user The weight matrix, Indicates the first Feature vectors of each user Represents items The weight matrix, Indicates the first The feature vector of an item.

Images