Interactive game data processing method and system applying virtual reality

Abstract

The embodiment of the invention discloses an interactive game data processing method and system applying virtual reality. The method comprises the steps that non-preset interactive action data which is generated by a user through VR interactive equipment and contains time sequence node marks are captured in real time; extracting action trend features, operation rhythm features and force fluctuation features in combination with time sequence relevance and spatial continuity to generate a behavior feature set; screening effective behavior features through an intention confidence set output by the interactive intention probability model; encoding the effective behavior characteristics according to a semantic mapping rule to generate game system semantic information containing interactive intention description; according to the type of the virtual scene, calling an adaptation rule, matching semantic information with a scene demand to determine a fit node, and adjusting parameters to generate a scene dynamic adaptation scheme; based on the adaptation scheme and the interaction intention description, feedback logic triggering conditions are adjusted, the real-time dynamic response of the virtual scene is updated, time sequence node marks are associated, and the understanding ability of the non-preset action intention is improved through the method.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an interactive game data processing method and system using virtual reality. Background Technology

[0002] With the rapid development of virtual reality technology, VR interactive games have become an important direction in the digital entertainment field. Existing VR game interaction systems typically rely on preset action libraries to achieve user interaction with virtual scenes. This involves matching user actions with predefined action templates to trigger corresponding scene feedback. To improve interaction accuracy, some existing technologies incorporate motion capture devices to collect real-time data such as the user's position, posture, and acceleration, using simple feature extraction to determine if the action conforms to the preset template. Simultaneously, to adapt to different scene types, existing technologies employ static scene rule libraries, pre-setting fixed interaction response logic for different game types such as racing and puzzle games. For example, in racing scenes, pre-set rules such as "steering actions trigger vehicle steering feedback" and "acceleration actions increase vehicle speed." Furthermore, in the feedback processing stage, existing technologies typically execute fixed response content based on preset trigger conditions, such as visually displaying steering animations or providing controller vibration feedback.

[0003] However, existing technologies are insufficient in processing non-preset interactive actions, cannot effectively identify impromptu operations by users due to sudden situations, and the scene adaptability and real-time performance of interactive feedback need to be improved. Static rule bases are difficult to dynamically match users' real interactive intentions and scene requirements. Summary of the Invention

[0004] This application provides a method and system for processing interactive game data using virtual reality, which enables accurate understanding of the intent of non-preset interactive actions and dynamic scene adaptation in virtual reality interactive games.

[0005] This application provides, in one aspect, a method for processing interactive game data using virtual reality, applied to an interactive game data processing system. The method includes: Real-time capture of non-preset interactive action data generated by users in a virtual scene through VR interactive devices, wherein the non-preset interactive action data includes time sequence node markers of action execution; By combining the temporal correlation of action execution with the continuity of spatial coordinate trajectory, behavioral features are extracted from the non-preset interactive action data to generate a set of behavioral features including action trend features, operation rhythm features, and intensity fluctuation features. The behavioral feature set is input into a pre-built interaction intent probability model, and the effective behavioral features are selected from the behavioral feature set using the intent confidence set output by the interaction intent probability model. The effective behavioral features are semantically encoded according to the semantic mapping rules preset by the game system to obtain a set of semantic elements. The semantic elements are then combined to generate semantic information of the game system that contains a description of the interactive intent. According to the type of the virtual scene, the matching scene adaptation rules are invoked. The semantic information of the game system and the scene adaptation rules are logically matched to determine the matching nodes between the semantic information of the game system and the interaction requirements of the virtual scene. Based on the matching nodes, the parameter distribution of the scene adaptation rules is adjusted and a dynamic scene adaptation scheme is generated. Based on the scene dynamic adaptation scheme and the interaction intent description, the feedback logic triggering conditions of the virtual scene are adjusted, and the real-time dynamic response of the virtual scene to the non-preset interaction action is updated according to the feedback logic triggering conditions. The interaction response timeliness of the real-time dynamic response is associated with the corresponding time sequence node marker.

[0006] One embodiment of this application provides an interactive game data processing system, including: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement any of the described interactive game data processing methods for virtual reality applications.

[0007] One embodiment of this application provides a readable storage medium storing a program or instructions, which, when executed by a processor, implements the steps of the interactive game data processing method for virtual reality.

[0008] Therefore, the embodiments of this application have the following beneficial effects: they achieve accurate intent understanding and dynamic scene adaptation for non-preset interactive actions in virtual reality interactive games. First, by capturing non-preset interactive action data containing temporal node markers in real time, and combining temporal correlation and spatial continuity to extract action trend features, operation rhythm features, and intensity fluctuation features, it breaks through the definition boundaries of interactive behavior in traditional preset action libraries, and can cover impromptu operations generated by users due to sudden scene situations. Second, by using an interaction intent probability model to perform confidence screening on the behavioral feature set, and through multi-branch collaborative calculation of intent confidence sets using temporal correlation, pattern matching, and anomaly detection, invalid action noise is effectively filtered, improving the system's accuracy in recognizing the user's true interaction intent. Furthermore, by using semantic mapping rules to transform effective behavioral features into game system semantic information containing interactive intent descriptions, a mapping bridge from action features to semantic intent is established, solving the semantic gap problem between unstructured action data and structured scene rules. Finally, based on the virtual scene type, adaptation rules are invoked, matching nodes are determined through logical matching, and parameter distribution is adjusted to generate a dynamic adaptation scheme. The interaction intent description is combined with the updated feedback logic trigger conditions and real-time dynamic response. The response time is associated with the time sequence node mark, which ensures the real-time performance and scene adaptability of the interaction feedback, enhances the immersion and personalization of virtual scene interaction, and provides richer interaction methods and a more intelligent feedback mechanism for virtual reality interactive games. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating an interactive game data processing method using virtual reality, provided as an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the basic structure of an interactive game data processing system provided in an embodiment of this application.

[0012] Figure 3 This is a functional block diagram of an interactive game data processing device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0014] Please see Figure 1 , Figure 1 This is a flowchart of an interactive game data processing method using virtual reality provided in an embodiment of this application. The method can be executed by an interactive game data processing system, or by the interactive game data processing system and a server. The method may include steps 110-160.

[0015] Step 110: Capture in real time the non-preset interactive action data generated by the user in the virtual scene through the VR interactive device, the non-preset interactive action data including the time sequence node markers of the action execution.

[0016] In this embodiment, the virtual scene is a virtual reality racing game scene, where the user wears a VR headset and holds a VR positioning controller to operate the game. The VR interaction device includes a VR positioning controller and a position tracking module built into the headset. The VR positioning controller outputs quaternion data of the controller posture, three-dimensional spatial coordinate data, and linear acceleration data. The position tracking module built into the headset outputs three-dimensional spatial coordinate data of the user's head. All data is transmitted to the interactive game data processing system in real time via a USB interface in JSON format. The system adds a timing node marker generated by the system clock to each data frame, with the timing node marker having millisecond precision. When the user suddenly swings the VR positioning controller quickly to the right due to a sudden situation on the track, this action is not in the game's preset interactive action library and is considered a non-preset interactive action. The system captures the corresponding controller posture quaternion data, three-dimensional spatial coordinate data, and linear acceleration data in real time and adds the corresponding timing node marker to the above data.

[0017] Step 120: Combine the temporal correlation of action execution with the continuity of spatial coordinate trajectory to extract behavioral features from the non-preset interactive action data, and generate a behavioral feature set including action trend features, operation rhythm features and force fluctuation features.

[0018] In this embodiment, for the non-preset interactive action data of the user rapidly waving the VR positioning handle to the right captured in step 110, the system first extracts the temporal correlation of the action execution, uses a sliding window to extract the action data subsequence, sets the sliding window length to 1 second and the step size to 0.2 seconds, and processes the extracted subsequence through a temporal convolutional network. The temporal convolutional network uses dilated causal convolution, and the dilation factor increases exponentially with the number of layers to capture multi-scale temporal dependencies. Then, the continuity of the spatial coordinate trajectory is extracted, and the three-dimensional spatial coordinate trajectory data of the VR positioning handle is smoothed by cubic spline interpolation. The curvature and velocity vector derivative of the trajectory in each frame are calculated. Finally, the temporal correlation features and spatial continuity features are fused to generate a set of behavioral features. The action trend features are the weighted average direction and displacement acceleration of the displacement vector within the window, the operation rhythm features are the main frequency components and energy distribution extracted by fast Fourier transform of the action speed sequence, and the force fluctuation features are the variance of the linear acceleration data of the VR positioning handle within the window and the impact peak exceeding twice the baseline threshold.

[0019] Step 130: Input the set of behavioral features into the pre-constructed interaction intent probability model, and use the intent confidence set output by the interaction intent probability model to filter out effective behavioral features from the set of behavioral features.

[0020] In this embodiment, the interaction intent probability model consists of a temporal correlation processing branch, a pattern matching processing branch, and a correlation anomaly detection processing branch. The behavioral feature set generated in step 120 is input into the model. The temporal correlation processing branch constructs a directed graph of action trends based on a graph neural network, where nodes represent action states at different time steps, edges represent state transitions, and it outputs the transition probabilities of each state and the first local intent confidence score. The pattern matching processing branch segments the operation rhythm features according to state labels and performs dynamic time warping to calculate similarity with a preset interaction rhythm template library, outputting the second local intent confidence score. The correlation anomaly detection processing branch uses statistical process control methods to identify abnormal fluctuation segments in the intensity fluctuation features, outputting the third local intent confidence score. The three local intent confidence scores are Min-Max normalized and weighted and fused according to preset weights of 0.4, 0.35, and 0.25. Combined with the feature correlation matrix, a comprehensive intent confidence score is calculated to form an intent confidence score set. The feature subset corresponding to the target intent confidence score exceeding the preset confidence threshold of 0.7 is taken as the effective behavioral features.

[0021] Step 131: Based on the constituent dimensions of the behavioral feature set, extract the common temporal node markers of each feature in the behavioral feature set as association anchors, perform feature dimension unification for the attribute differences of each feature, and generate a standardized behavioral feature subset containing temporal anchors.

[0022] In this embodiment, the behavioral feature set includes action trend features, operation rhythm features, and intensity fluctuation features. The common temporal nodes of each feature are marked as timestamps during the action execution process, and these timestamps are extracted as association anchors. Action trend features are three-dimensional vectors, operation rhythm features are one-dimensional scalars, and intensity fluctuation features are two-dimensional vectors. To address the differences in these attributes, a feature scaling method is used to map each feature to the same numerical range, generating a standardized subset of behavioral features containing temporal anchors.

[0023] Step 132: Import the action trend features and corresponding temporal anchors into the preset temporal association processing branch in the interaction intent probability model. Extract the temporal dependencies of features at different time granularities through a variable step-size sliding window and construct a directed action trend graph. Calculate the occurrence probability of each transition path in the directed action trend graph and map it to the first intent association coefficient based on the transition time. Generate the first local intent confidence corresponding to the action trend features based on the coefficient distribution and the temporal continuity of the temporal anchors. Simultaneously output the trend status label corresponding to each temporal anchor.

[0024] In this embodiment, the step size of the variable step-size sliding window is dynamically adjusted according to the action speed; the step size increases when the action speed is fast and decreases when the action speed is slow. The variable step-size sliding window extracts the temporal dependencies of action trend features at different time granularities, constructing a directed graph of action trends. Nodes in the graph represent action trend feature states, and edges represent state transition directions. The probability of occurrence of each transition path is calculated, and combined with the inverse weighted mapping of transition time (timestamp difference) to obtain the first intent association coefficient. Based on the coefficient distribution and the temporal continuity of the time-series anchor points, a first local intent confidence score is generated, and the trend state labels corresponding to each time-series anchor point are output synchronously, such as "accelerate turning" and "uniform speed movement".

[0025] Step 133: Call the pattern matching processing branch of the interaction intent probability model to receive the temporal anchor points and trend status labels output by the temporal association processing branch. Segment the operation rhythm feature into segment rhythm features according to the temporal anchor points. Perform periodic splitting on each segment rhythm feature and perform targeted comparison with the preset interaction rhythm template library to obtain segment similarity. Convert the weighted average of each segment similarity into a second intent association coefficient. Combine the intent labels of the preset interaction rhythm template library and the association anchor points to generate the second local intent confidence corresponding to the operation rhythm feature. Simultaneously output the matching degree between the segment rhythm feature and the trend status label.

[0026] In this embodiment, the preset interaction rhythm template library includes templates such as "continuous rapid clicks" and "slow waving". The operation rhythm features are divided into multiple segmented rhythm features according to time-series anchor points. Periodic segmentation is performed on each segmented rhythm feature. The segment similarity is obtained by directional comparison with the template library through a dynamic time warping algorithm. The weighted average of the segment similarities is converted into a second intent association coefficient. The second local intent confidence is generated by combining the intent labels and associated anchor points of the template library. The matching degree between the segmented rhythm features and the trend state labels is output simultaneously, such as "acceleration state and fast rhythm matching degree 0.9".

[0027] Step 134: Activate the correlation anomaly detection processing branch of the interaction intent probability model, import the intensity fluctuation feature, time sequence anchor point, trend state label and segmented rhythm feature, determine the intensity benchmark range of different time sequence stages based on the trend state label, adjust the dynamic floating threshold of the benchmark range in combination with the periodic characteristics of the segmented rhythm feature, identify the fluctuation segment in the intensity fluctuation feature that exceeds the dynamic benchmark range, quantify the effective interaction intensity change judgment result of the fluctuation segment and the synchronization coefficient of the segmented rhythm feature into a third intent correlation coefficient, generate the third local intent confidence corresponding to the intensity fluctuation feature, and output the correlation analysis result of the fluctuation segment with the operation rhythm feature and the action trend feature.

[0028] In this embodiment, the action execution process is divided into "start-up phase", "stable phase" and "end phase" based on trend status labels. Each phase sets a strength benchmark range based on historical data. The dynamic floating threshold of the benchmark range is adjusted in combination with the cycle length of the segmented rhythm features. If the cycle length is short, the threshold is tightened. Fluctuation segments in the strength fluctuation features that exceed the dynamic benchmark range are identified. The effective interaction strength change judgment result of the fluctuation segment and the synchronization coefficient of the segmented rhythm features are quantified into a third intention correlation coefficient to generate a third local intention confidence. The correlation analysis results of the fluctuation segment and the operation rhythm features and action trend features are output, such as "abnormal force exertion and turning trend synchronization".

[0029] Step 135: Using the associated anchor point as an index, standardize the first local intent confidence, the second local intent confidence, and the third local intent confidence. Weight and fuse the standardized local intent confidences according to preset weights to obtain a fused confidence. Quantify the trend status label, the matching fit, and the association analysis result into numerical features. Combine these numerical features based on preset association rules to generate a quantified feature association matrix. Calculate the comprehensive intent confidence corresponding to each feature subset within the standardized behavioral feature subset based on the quantified values ​​of each element in the feature association matrix and the fused confidence, forming an intent confidence set. Select the feature subsets in the intent confidence set that exceed the target intent confidence level as the effective behavioral features.

[0030] In this embodiment, Min-Max normalization is used to standardize the three local intent confidence scores. The preset weights are 0.4 for the temporal association processing branch, 0.35 for the pattern matching processing branch, and 0.25 for the association anomaly detection processing branch. The standardized local intent confidence scores are then weighted and fused to obtain the fused confidence score. The trend status labels are one-hot encoded into vectors, and the matching fit is directly used as a scalar. Relationships such as "synchronous" and "asynchronous" in the association analysis results are quantized as 1 / 0. A feature association matrix is ​​generated based on preset association rules. According to the quantized values ​​of each element in the feature association matrix and the fused confidence score, the comprehensive intent confidence score is calculated using the Sigmoid function mapping, forming an intent confidence score set. A preset confidence threshold of 0.7 is set, and the feature subset corresponding to the target intent confidence score exceeding this threshold is considered as effective behavioral features.

[0031] Step 140: Perform semantic encoding processing on the effective behavioral features according to the semantic mapping rules preset by the game system to obtain a set of semantic elements, and generate game system semantic information containing interactive intent descriptions through the association and combination of the set of semantic elements.

[0032] In this embodiment, the game system's preset semantic mapping rule base stores the matching relationships between key feature items and candidate semantic factors. Key feature items are extracted from effective behavioral features, such as "turning angular velocity greater than 30° / s" and "main frequency between 2-4Hz," and matched with the rule base to obtain candidate semantic factors. Candidate semantic factors are sorted by temporal anchor points, and the semantic association strength between adjacent factors is calculated. Isolated factors below a threshold of 0.6 are removed to obtain core semantic factors. Modifying attributes are added to the core semantic factors, such as combining "turning" with "angular velocity value = 35" to enhance it into "rapid turning." A semantic network is constructed using factors as nodes and association strength as edge weights. Potential associations are discovered and semantic links are supplemented using graph mining algorithms. Based on the semantic network topology, strongly connected semantic elements are combined to generate an interaction intent description, such as "the user intends to make a rapid right turn, followed by continuous clicking at a frequency of approximately 3Hz," integrating these elements to generate the game system's semantic information.

[0033] Step 141: Extract key feature items with intent orientation from each feature of the effective behavioral features, match the key feature items with the semantic mapping rules preset by the game system, and obtain the candidate semantic factors corresponding to each key feature item.

[0034] In this embodiment, the effective behavioral features include action trend features, operation rhythm features, and force fluctuation features. From the action trend features, "turning angular velocity greater than 30° / s" is extracted as a key feature item; from the operation rhythm features, "main frequency between 2-4Hz" is extracted as a key feature item; and from the force fluctuation features, "acceleration variance greater than 5" is extracted as a key feature item. In the game system's preset semantic mapping rule base, "turning angular velocity greater than 30° / s" matches the candidate semantic factor "rapid turning," "main frequency between 2-4Hz" matches the candidate semantic factor "continuous operation," and "acceleration variance greater than 5" matches the candidate semantic factor "large force variation," thus obtaining the candidate semantic factor corresponding to each key feature item.

[0035] Step 142: Based on the temporal anchor points of each feature in the effective behavioral features, sort the candidate semantic factors in a temporal order to obtain a candidate semantic factor sequence; calculate the semantic association strength quantification value between adjacent semantic factors in the candidate semantic factor sequence, and according to the preset association strength threshold, remove isolated candidate semantic factors whose semantic association strength quantification value is lower than the association strength threshold to obtain core semantic factors with continuous semantic association.

[0036] In this embodiment, the temporal anchor point of each feature in the effective behavioral features is the timestamp during the action execution process. The candidate semantic factors "rapid turning", "continuous operation", and "large change in intensity" are sorted according to the order of the timestamps to obtain a candidate semantic factor sequence. The cosine similarity algorithm is used to calculate the semantic association strength quantification value between adjacent semantic factors. The preset association strength threshold is 0.6. If the semantic association strength quantification value between "rapid turning" and "continuous operation" is 0.8, and the semantic association strength quantification value between "continuous operation" and "large change in intensity" is 0.7, both of which are higher than the threshold, all candidate semantic factors are retained as core semantic factors. If the semantic association strength quantification value of an adjacent factor is lower than 0.6, the isolated candidate semantic factor is removed.

[0037] Step 143: Perform semantic attribute enhancement processing on the core semantic factors, add semantic modification attributes to each core semantic factor in combination with the feature attributes of the effective behavioral features, and associate the core semantic factors with the corresponding effective behavioral features through the semantic modification attributes to generate semantic elements with attribute identifiers.

[0038] In this embodiment, the core semantic factors are "rapid steering," "continuous operation," and "large force variation." The characteristic attributes of the effective behavioral features include steering angular velocity value, operation frequency, and acceleration variance. The semantic modifier attribute "steering angular velocity value = 35° / s" is added to "rapid steering," "operation frequency = 3Hz" is added to "continuous operation," and "acceleration variance = 6" is added to "large force variation." These semantic modifier attributes associate the core semantic factors with the corresponding effective behavioral features, generating semantic elements with attribute identifiers, such as "rapid steering (steering angular velocity value = 35° / s)," "continuous operation (operation frequency = 3Hz)," and "large force variation (acceleration variance = 6)."

[0039] Step 144: Construct a semantic element network by arranging the semantic elements with attribute identifiers according to temporal anchor points and semantic association strength. In the semantic element network, mine the potential associations between semantic elements and supplement the semantic connection links based on the potential associations to obtain a set of semantic elements with logical associations.

[0040] In this embodiment, semantic elements with attribute identifiers are sorted by temporal anchor points, and a semantic element network is constructed using semantic elements as nodes and semantic association strength as edge weights. A community detection algorithm is used to mine potential associations between semantic elements in the network. If "fast turn" and "continuous operation" are not adjacent but belong to the same community and have a high historical co-occurrence frequency, a semantic connection link is added between them. By mining potential associations and adding semantic connection links, a set of logically related semantic elements is obtained.

[0041] Step 145: Based on the semantic modification attributes and semantic associations of each semantic element in the semantic element set, perform the association and combination operation of the semantic elements to generate an interaction intent description that represents the user's interaction purpose, and integrate the interaction intent description with the associated semantic element set to generate game system semantic information.

[0042] In this embodiment, the semantic element set includes semantic elements such as "rapid turning (turning angular velocity = 35° / s)," "continuous operation (operation frequency = 3Hz)," and "large force variation (acceleration variance = 6)." The semantic modification attributes of each semantic element include turning angular velocity, operation frequency, and acceleration variance. The semantic association is "rapid turning" followed by "continuous operation" and accompanied by "large force variation." Based on the above semantic modification attributes and semantic association, an associated combination operation is performed to generate an interaction intent description: "The user intends to perform a rapid right turn with a turning angular velocity of 35° / s, followed by continuous operation at a frequency of 3Hz. During the operation, the force variation is large, and the acceleration variance is 6." This interaction intent description is then integrated with the associated semantic element set to generate the game system semantic information.

[0043] Step 150: Invoke the matching scene adaptation rules according to the type of the virtual scene, logically match the game system semantic information and the scene adaptation rules to determine the matching nodes between the game system semantic information and the interaction requirements of the virtual scene, adjust the parameter distribution of the scene adaptation rules based on the matching nodes and generate a dynamic scene adaptation scheme.

[0044] In this embodiment, the virtual scene type is racing. Matching scene adaptation rules are retrieved from the scene rule library. These rules are stored in XML format and contain a description of scene interaction requirements. A sentence vector model is used to encode the interaction intent description into a vector. Core attributes from the semantic element set are extracted as feature vectors. A mapping model from intent vectors to semantic attribute vectors is established using a shallow neural network. The cosine similarity between the intent vector and the scene interaction requirement description vector is calculated. Interaction intents exceeding a threshold of 0.8 and satisfying logical constraints are identified and marked as initial matching nodes. The rule execution logic is used to determine whether the semantic element corresponding to the initial matching node triggers the rule execution process, thus identifying the matching node. Based on the matching node, associated parameter items are located in the parameter association network. The distribution of parameter item values ​​is adjusted according to the intent association weights of the semantic elements. A dynamic scene adaptation scheme is generated based on the adjusted parameter distribution, the matching node, and the scene adaptation rules.

[0045] Step 151: Extract the type features of the virtual scene, match the type features with the preset scene rule library, and call the scene adaptation rules that are compatible with the type of the virtual scene.

[0046] In this embodiment, the virtual scene is a virtual reality racing game scene, and its type characteristics include "racing," "track scene," and "vehicle driving interaction." A preset scene rule base stores scene adaptation rules corresponding to different types of virtual scenes. Each rule is stored in XML format and includes a rule identifier, scene type matching features, scene interaction requirement description, parameter configuration, etc. The type characteristics of the virtual scene are matched against the scene rule base to find a matching scene adaptation rule, such as "racing track scene vehicle driving interaction adaptation rule," and this rule is called for subsequent logical matching.

[0047] Step 152: Convert the interaction intent description into a standardized intent expression vector, extract the core semantic attributes related to scene interaction from the semantic element set, and establish an intent semantic association model. The intent semantic association model reflects the correspondence between interaction intent and semantic elements.

[0048] In this embodiment, the interaction intent is described as "the user intends to make a quick right turn, followed by continuous clicking at a frequency of approximately 3Hz." This intent is converted into a standardized intent expression vector using a sentence vector model (such as BERT), with a fixed vector dimension. Core semantic attributes related to the scene interaction, such as "turning direction," "turning speed," and "operation frequency," are extracted from the semantic element set to establish an intent semantic association model. This model takes the intent expression vector as input and the core semantic attributes of the semantic elements as output, and is implemented through a shallow neural network to reflect the correspondence between the interaction intent and the semantic elements.

[0049] Step 153: Logically match the intent semantic association model with the scene interaction requirement description in the scene adaptation rules, and use the semantic similarity comparison method to calculate the semantic similarity quantification value between the interaction intent and the scene interaction requirement; based on the preset similarity threshold, identify the interaction intents whose semantic similarity quantification value exceeds the threshold and meet the logical constraints of the scene interaction requirement, and mark the logical position of the interaction intent that meets the conditions in the intent semantic association model and the scene adaptation rules as the initial matching node.

[0050] In this embodiment, the scene interaction requirement in the scene adaptation rule is described as "In racing scenarios, when a user makes a quick turn, vehicle steering feedback and track prompts need to be triggered." A cosine similarity algorithm is used to calculate the semantic similarity metric between the interaction intent expression vector and the scene interaction requirement description vector. A preset similarity threshold is 0.8. If the calculated semantic similarity metric is 0.85, exceeding the threshold and satisfying the scene interaction requirement logical constraints (e.g., "quick turn" meets the requirement of "user quick turn"), then the output node corresponding to the interaction intent in the intent semantic association model and the requirement description node corresponding to it in the scene adaptation rule are marked as initial matching nodes.

[0051] Step 154: Combine the rule execution logic of the scene adaptation rule to determine whether the semantic element corresponding to the initial matching node triggers the rule execution process. Determine the matching node based on the trigger judgment result. Using the matching node as a benchmark, identify the parameter items associated with the matching node based on the parameter distribution of the scene adaptation rule. Adjust the value distribution of the parameter items according to the intent association weight of the semantic elements in the semantic information of the game system.

[0052] In this embodiment, the rule execution logic of the scene adaptation rule is: "If the semantic element corresponding to the initial matching node contains 'quick turn,' then the vehicle steering feedback rule execution process is triggered." The rule execution process is triggered by determining that the semantic element "quick turn" corresponding to the initial matching node triggers the rule execution process, thus identifying the initial matching node as the matching node. Based on the matching node, associated parameter items are located in the parameter association network of the scene adaptation rule, such as "steering feedback intensity parameter" and "track prompt display duration parameter." The intent association weight of the semantic element "quick turn" in the game system semantic information is 0.8. The parameter value distribution is adjusted according to this weight; for example, the steering feedback intensity parameter value is adjusted from 0.6 to 0.8, and the track prompt display duration parameter value is adjusted from 2 seconds to 3 seconds.

[0053] Step 1541: Using parameter items as nodes and the mutual influence relationships between parameter items as edges, locate the core rule entry corresponding to each matching node in the parameter association network of the scene adaptation rule, track all target parameter items involved in the core rule entry, and mark the target parameter items as candidate association parameters.

[0054] In this embodiment, the parameter association network of the scene adaptation rule uses parameter items as nodes and the mutual influence relationships between parameter items as edges. For example, "steering feedback intensity parameter" and "vehicle steering sensitivity parameter" have a mutual influence relationship. The core rule entry corresponding to the matching node is the "vehicle steering feedback rule". The target parameter items involved in this core rule entry are tracked, including "steering feedback intensity parameter", "vehicle steering sensitivity parameter", and "steering feedback delay parameter", and the above target parameter items are marked as candidate association parameters.

[0055] Step 1542: Based on the quantified value of the influence of parameters on the rule execution effect and the quantified value of the matching between parameters and semantic elements, calculate the quantified value of the association strength between each candidate association parameter and the matching node according to the preset association strength calculation rules; select the parameter items whose association strength quantified value exceeds the preset strength threshold as key parameter items associated with the matching node, and establish a correspondence table between key parameter items and matching nodes.

[0056] In this embodiment, the quantified value of the influence of parameters on the rule execution effect is obtained through historical A / B test data. For example, the quantified value of the influence of the "steering feedback intensity parameter" is 0.7, the "vehicle steering sensitivity parameter" is 0.6, and the "steering feedback delay parameter" is 0.5. The quantified value of the matching between parameters and semantic elements is calculated by the text similarity between the parameter description and the core attributes of the semantic element. For example, the quantified value of the matching between the "steering feedback intensity parameter" and the semantic element "fast steering" is 0.8, the "vehicle steering sensitivity parameter" is 0.7, and the "steering feedback delay parameter" is 0.6. The preset association strength calculation rule is: association strength quantified value = influence quantified value × 0.6 + matching quantified value × 0.4. The calculated association strength quantified values ​​are: 0.7 × 0.6 + 0.8 × 0.4 = 0.74 for the "steering feedback intensity parameter", 0.6 × 0.6 + 0.7 × 0.4 = 0.64 for the "vehicle steering sensitivity parameter", and 0.5 × 0.6 + 0.6 × 0.4 = 0.54 for the "steering feedback delay parameter". With a preset strength threshold of 0.6, "steering feedback strength parameter" and "vehicle steering sensitivity parameter" were selected as key parameters, and a correspondence table between key parameters and matching nodes was established.

[0057] Step 1543: Extract the intent association weights of each semantic element in the semantic information of the game system, normalize the intent association weights to obtain standardized weight values, match the standardized weight values ​​with the key parameter items in the corresponding relationship table, and determine the weight reference value corresponding to each key parameter item. The weight reference value reflects the degree of guidance of the interaction intent on parameter adjustment.

[0058] In this embodiment, the semantic elements in the game system semantic information include "rapid steering," "continuous operation," and "large force change," with intent association weights of 0.8, 0.5, and 0.3, respectively. These weights are normalized by dividing each weight by the sum of its weights, i.e., 0.8 + 0.5 + 0.3 = 1.6. Therefore, the normalized weight value for "rapid steering" is 0.8 / 1.6 = 0.5, for "continuous operation" it is 0.5 / 1.6 = 0.3125, and for "large force change" it is 0.3 / 1.6 = 0.1875. The key parameter items in the correspondence table are "steering feedback intensity parameter" and "vehicle steering sensitivity parameter." The normalized weight values ​​are matched with the key parameter items; the "steering feedback intensity parameter" corresponds to a normalized weight value of 0.5 for "rapid steering," and the "vehicle steering sensitivity parameter" corresponds to a normalized weight value of 0.5 for "rapid steering." Therefore, the reference weight value for each key parameter item is determined to be 0.5.

[0059] Step 1544: Obtain the current value and adjustment range of each key parameter item. Using the weight reference value as a guide, generate a parameter adjustment strategy based on the mutual constraints between parameters. Perform multiple rounds of parameter adjustment operations according to the parameter adjustment strategy until the adjusted parameter distribution matches the requirements for realizing the interaction intent.

[0060] In this embodiment, the current value of the "steering feedback intensity parameter" is 0.6, with an adjustment range of 0.4-0.9; the current value of the "vehicle steering sensitivity parameter" is 0.5, with an adjustment range of 0.3-0.8. The weight reference value is 0.5, and the mutual constraint relationship between the parameters is "steering feedback intensity parameter + vehicle steering sensitivity parameter ≤ 1.5". The parameter adjustment strategy is generated using the weight reference value as a guide. For example, in the first round of adjustment, the value of the "steering feedback intensity parameter" is increased by 0.1 to 0.7, and the value of the "vehicle steering sensitivity parameter" is increased by 0.1 to 0.6. The constraint relationship 0.7 + 0.6 = 1.3 ≤ 1.5 is checked and is satisfied. In the second round of adjustment, the values ​​are further increased until the adjusted parameter distribution matches the requirements for realizing the interaction intent. For example, the value of the "steering feedback intensity parameter" is 0.8, and the value of the "vehicle steering sensitivity parameter" is 0.7, satisfying the constraint and matching the requirements for realizing the interaction intent.

[0061] Step 155: Generate a dynamic scene adaptation scheme based on the adjusted parameter distribution, the matching nodes, and the scene adaptation rules.

[0062] In this embodiment, the adjusted parameter distribution includes a "steering feedback intensity parameter" value of 0.8, a "vehicle steering sensitivity parameter" value of 0.7, and a "track prompt display duration parameter" value of 3 seconds. The matching node is the logical position corresponding to "rapid steering," and the scene adaptation rule is the "vehicle driving interaction adaptation rule for racing track scenes." Based on the above content, a scene dynamic adaptation scheme is generated. The scheme includes adaptation rule identifiers, matching node information, adjusted parameter configurations, and rule execution logic adjustment instructions, which are used to guide subsequent virtual scene feedback logic adjustments and real-time dynamic response updates.

[0063] Step 160: Adjust the feedback logic triggering conditions of the virtual scene based on the scene dynamic adaptation scheme and the interaction intent description, and update the real-time dynamic response of the virtual scene to the non-preset interaction action according to the feedback logic triggering conditions. The interaction response timeliness of the real-time dynamic response is associated with the corresponding time sequence node marker.

[0064] In this embodiment, based on the interactive intent description, the rules, constraints, and parameter adjustment directions related to feedback in the scene dynamic adaptation scheme are analyzed to generate a priority sequence for feedback logic adjustment. The constituent elements of the current feedback logic triggering condition are extracted, and mismatched logic nodes are identified as optimization targets. A case-based reasoning system is invoked to retrieve the historical case library and generate a triggering condition adjustment strategy. The original triggering condition logic expression is modified, and an association mapping with the time sequence node marker is established. A multi-dimensional dynamic response candidate set is generated by combining scene attributes and real-time status. A two-layer neural network is used to calculate semantic relevance and scene adaptability to select the optimal dynamic response scheme. Real-time interactive feedback is updated according to the optimal scheme, and the response execution information and time sequence node marker are synchronously associated and stored.

[0065] Step 161: Based on the interaction intent description, analyze the rule constraints and parameter adjustment directions related to virtual scene feedback in the scene dynamic adaptation scheme, and transform the core demands of the interaction intent into a priority sequence of feedback logic adjustment, so that the optimization direction of the feedback logic is consistent with the user's interaction purpose.

[0066] In this embodiment, the interaction intent is described as "the user intends to make a quick right turn, followed by continuous clicking at a frequency of approximately 3Hz," with the core requirements being "quick right turn feedback" and "continuous click feedback." The rules and constraints related to virtual scene feedback in the dynamic scene adaptation scheme are analyzed, such as "turn feedback must respond within 50ms after the action is triggered" and "continuous click feedback must match the operation frequency." The parameter adjustment direction is "increasing the intensity of turn feedback" and "adjusting the sensitivity of continuous click feedback." The core requirements of the interaction intent are transformed into a priority sequence for feedback logic adjustment, with priorities from high to low: "adjusting quick right turn feedback," "adjusting continuous click feedback," and "optimizing feedback intensity parameters," ensuring that the optimization direction of the feedback logic is consistent with the user's interaction purpose.

[0067] Step 162: Based on the priority sequence, extract the constituent elements of the current feedback logic triggering conditions of the virtual scene, and combine them with the adjusted parameter distribution in the scene dynamic adaptation scheme to identify the logic nodes in the current feedback logic triggering conditions that do not match the scene adaptation requirements and interaction intentions, and obtain the set of triggering condition optimization target points.

[0068] In this embodiment, the priority sequence is "fast right turn feedback adjustment," "continuous click feedback adjustment," and "feedback intensity parameter optimization." The components of the current feedback logic triggering conditions in the virtual scene include "steering angular velocity threshold," "operation frequency threshold," and "feedback intensity parameter." The adjusted parameter distribution in the scene dynamic adaptation scheme is "steering angular velocity threshold adjusted to 25° / s," "operation frequency threshold adjusted to 2Hz," and "feedback intensity parameter adjusted to 0.8." The current feedback logic triggering conditions, including "steering angular velocity threshold of 30° / s," "operation frequency threshold of 3Hz," and "feedback intensity parameter of 0.6," do not match the adjusted parameter distribution and the core requirements of the interaction intent. Therefore, these logical nodes are identified as target points for triggering condition optimization, resulting in a set of target points for triggering condition optimization.

[0069] Step 163: Call the preset feedback logic reasoning model to perform logical association reasoning on the set of target points for trigger condition optimization, the description of interaction intent, and the scene dynamic adaptation parameters to generate a trigger condition adjustment strategy.

[0070] In this embodiment, the preset feedback logic reasoning model is a case-based reasoning system. The case library is stored in a graph structure, with nodes representing adjustment strategies and edges representing scene similarity. The system inputs a set of target points for trigger condition optimization, an interaction intent description, and scene dynamic adaptation parameters. It then retrieves adjustment strategies for similar scenes from the case library, such as "adjusting the steering angular velocity threshold to 25° / s," "adjusting the operation frequency threshold to 2Hz," and "adjusting the feedback intensity parameter to 0.8." The strategy of the most similar case is output as the trigger condition adjustment strategy.

[0071] Step 1631: Extract the logical attributes, parameter constraints, and association levels with historical feedback logic of each optimization target in the set of optimization targets for triggering conditions. After distinguishing core targets and associated targets according to the association level, construct a target feature matrix based on target type, attribute features, and association weight.

[0072] In this embodiment, the set of target points for trigger condition optimization includes "steering angular velocity threshold mismatch", "operation frequency threshold mismatch", and "feedback intensity parameter mismatch". The logical attributes of each optimization target point are extracted, such as "steering angular velocity threshold mismatch" being a "threshold type", "operation frequency threshold mismatch" being a "threshold type", and "feedback intensity parameter mismatch" being a "parameter type"; parameter constraints include "steering angular velocity threshold adjustment range 0.4-0.9", "operation frequency threshold adjustment range 0.3-0.8", and "feedback intensity parameter adjustment range 0.4-0.9"; the association levels with historical feedback logic are as follows: "steering angular velocity threshold mismatch" is associated at level 1 (core level), "operation frequency threshold mismatch" is associated at level 2 (association level), and "feedback intensity parameter mismatch" is associated at level 3 (association level). Based on the association level, the core target point is "steering angular velocity threshold mismatch", and the associated targets are "operation frequency threshold mismatch" and "feedback intensity parameter mismatch". Construct a target feature matrix based on target type, attribute features, and association weights. Rows in the matrix represent targets, and columns represent feature dimensions, such as target type, parameter constraints, association level, and association weights.

[0073] Step 1632: Extract the core demand items, scenario adaptation expectations, and response timeliness requirements of the interaction intent description, map the core demand items into quantitative intent indicators, transform the scenario adaptation expectations into scenario constraint parameters, and generate timeliness thresholds by associating time sequence node markers with response timeliness requirements to obtain the intent quantification feature vector.

[0074] In this embodiment, the core requirements of the interaction intent description are "quick right turn" and "continuous clicks," which are mapped to quantified intent indicators. For example, "quick right turn" is mapped to a "turning speed quantification value of 0.8," and "continuous clicks" is mapped to an "operation frequency quantification value of 0.7." The scenario adaptation expectation is "smooth steering feedback" and "synchronous continuous click feedback," which are converted into scenario constraint parameters, such as "steering feedback transition time < 0.1s" and "continuous click feedback delay < 30ms." The response timeliness requirement is "feedback must respond within 50ms after the action is triggered," and a timeliness threshold is generated by associating time sequence node markers with the time sequence node markers, i.e., 50ms after the timestamp corresponding to the time sequence node marker. The above content is integrated to obtain an intent quantification feature vector, and the vector dimensions include quantified intent indicators, scenario constraint parameters, and timeliness thresholds.

[0075] Step 1633: Classify the scene dynamic adaptation parameters into scene basic parameters, interaction response parameters, and state adjustment parameters. Calculate the deviation between the current value and the optimal range of each type of parameter. Combine the influence weight of the parameter on the feedback logic to generate the parameter influence coefficient and obtain the parameter feature set.

[0076] In this embodiment, the scene dynamic adaptation parameters include "steering feedback intensity parameter," "vehicle steering sensitivity parameter," "track prompt display duration parameter," and "operation frequency response threshold parameter," etc. Classified by scene basic parameters, interaction response parameters, and state adjustment parameters, "track prompt display duration parameter" is a scene basic parameter, while "steering feedback intensity parameter," "vehicle steering sensitivity parameter," and "operation frequency response threshold parameter" are interaction response parameters. The deviation between the current value and the optimal range of each parameter is calculated. For example, if the current value of the "steering feedback intensity parameter" is 0.6, and the optimal range is 0.4-0.9, the deviation = (current value - midpoint of optimal range) / range radius = (0.6-0.65) / 0.25 = -0.2; if the current value of the "vehicle steering sensitivity parameter" is 0.5, and the optimal range is 0.3-0.8, the deviation = (0.5-0.55) / 0.25 = -0.2; if the current value of the "operation frequency response threshold parameter" is 3Hz, and the optimal range is 2-4Hz, the deviation = (3-3) / 1 = 0. The weights of the parameters' influence on the feedback logic are set using an expert scoring method. For example, the influence weight of the "steering feedback intensity parameter" is 0.7, the "vehicle steering sensitivity parameter" is 0.6, and the "operation frequency response threshold parameter" is 0.5. The parameter influence coefficient is generated as deviation × influence weight. For example, the influence coefficient of the "steering feedback intensity parameter" is -0.2 × 0.7 = -0.14, the "vehicle steering sensitivity parameter" is -0.2 × 0.6 = -0.12, and the "operation frequency response threshold parameter" is 0 × 0.5 = 0, thus obtaining the parameter feature set.

[0077] Step 1634: Align the target feature matrix, the intent quantification feature vector, and the parameter feature set into the input feedback logic reasoning model, and assign the highest weight to the core target features and the core intent appeal features, and assign secondary weight to the associated target features and parameter influence coefficients.

[0078] In this embodiment, the feedback logic reasoning model is a case-based reasoning system. The target feature matrix, intent quantification feature vector, and parameter feature set are input into the model and aligned. The alignment method is matching by feature dimension. For example, "threshold-type" targets in the target feature matrix are matched with "quantified intent indicators" in the intent quantification feature vector, and "interaction response parameters" in the parameter feature set are matched with "parameter-type" targets in the target feature matrix. The core target features (such as "steering angular velocity threshold mismatch") and core intent appeal features (such as "rapid right turn") are assigned the highest weight of 0.7, while related target features (such as "operation frequency threshold mismatch" and "feedback intensity parameter mismatch") and parameter influence coefficients are assigned a secondary weight of 0.3.

[0079] Step 1635: Using the feedback logic reasoning model, establish the relationship between features based on predicate logic rules, calculate the matching degree between the target optimization requirement features and the intention quantitative indicator features, define the feasibility boundary of target adjustment in combination with parameter influence coefficients, and generate multiple preliminary logical reasoning paths with quantitative matching degree evaluation.

[0080] In this embodiment, the feedback logic reasoning model incorporates a predicate logic rule library, such as "IF target type = threshold type AND intent index = high steering speed quantification value AND parameter influence coefficient < 0 THEN, it is recommended to lower the threshold" and "IF target type = parameter type AND intent index = high operation frequency quantification value AND parameter influence coefficient = 0 THEN, it is recommended to maintain the parameter value". The target optimization requirement characteristics, intent quantification index characteristics, and parameter influence coefficients are input into the model. Based on the predicate logic rules, a correlation is established, and the matching degree is calculated. For example, the matching degree between the target point "steering angle velocity threshold mismatch" and the intent index "rapid right turn" is 0.8; the matching degree between the target point "operation frequency threshold mismatch" and the intent index "continuous click" is 0.7; and the matching degree between the target point "feedback intensity parameter mismatch" and the intent index "feedback intensity optimization" is 0.6. The feasibility boundary of target point adjustment is defined by combining the parameter influence coefficient, such as "the adjustment range of steering angular velocity threshold is 0.4-0.9", and a preliminary logical reasoning path is generated. Each path is accompanied by a quantitative matching degree evaluation, such as "reducing the steering angular velocity threshold to 25° / s, matching degree 0.8", "reducing the operating frequency threshold to 2Hz, matching degree 0.7", and "increasing the feedback intensity parameter to 0.8, matching degree 0.6".

[0081] Step 1636: Using the historical interaction data and feedback logic execution cases of the virtual scene, evaluate and filter each inference path based on the execution success rate quantification value and the interaction adaptation quantification value; perform deduplication optimization on the effective inference paths remaining after filtering; extract the target adjustment direction, parameter modification range and logical association rules from the deduplicated and optimized inference paths, and determine the execution order of each strategy based on the core target adjustment strategy and the associated target adjustment strategy, combined with the timeliness requirements in the intent quantification feature vector, and generate the trigger condition adjustment strategy.

[0082] In this embodiment, historical interaction data and feedback logic execution cases of the virtual scene are stored in a case library. Each case includes information such as inference path, execution success rate, and interaction adaptability. The initial logic inference paths are evaluated: "reducing the steering angular velocity threshold to 25° / s" has an execution success rate of 0.9 and an interaction adaptability of 0.85; "reducing the operation frequency threshold to 2Hz" has an execution success rate of 0.85 and an interaction adaptability of 0.8; and "increasing the feedback intensity parameter to 0.8" has an execution success rate of 0.8 and an interaction adaptability of 0.75. A filtering condition is set as execution success rate > 0.8 and interaction adaptability > 0.75, retaining the first two paths after filtering. Deduplication optimization is performed on the retained paths, ensuring no duplicate paths. The target adjustment direction, parameter modification range, and logical association rules are extracted from the deduplicated and optimized inference path. Based on the adjustment strategy of the core target "steering angle velocity threshold mismatch" and supplemented by the adjustment strategy of the associated target "operation frequency threshold mismatch", combined with the timeliness requirement in the intent quantification feature vector "feedback must respond within 50ms after the action is triggered", the strategy execution order is determined to be "first reduce the steering angle velocity threshold to 25° / s, then reduce the operation frequency threshold to 2Hz", and the trigger condition adjustment strategy is generated.

[0083] Step 164: Determine the feedback logic triggering reconstruction condition of the virtual scene according to the triggering condition adjustment strategy, establish the association mapping between the feedback logic triggering reconstruction condition and the time sequence node marker in the non-preset interactive action data, so that the activation timing of the feedback logic triggering reconstruction condition matches the time characteristics of the action execution.

[0084] In this embodiment, the trigger condition adjustment strategy is to "reduce the steering angular velocity threshold to 25° / s and reduce the operation frequency threshold to 2Hz". The feedback logic trigger reconstruction condition for the virtual scene is determined to be "steering angular velocity > 25° / s and operation frequency > 2Hz". In the non-preset interactive action data, the time sequence nodes are marked as timestamps during the action execution process. An association mapping is established between the feedback logic trigger reconstruction condition and the time sequence node markings. For example, the "steering angular velocity > 25° / s" condition is bound to the steering angular velocity data stream of the corresponding timestamp, and the "operation frequency > 2Hz" condition is bound to the operation frequency data stream of the corresponding timestamp, so that the activation timing of the trigger reconstruction condition matches the time characteristics of the action execution.

[0085] Step 165: Based on the feedback logic triggering the reconstruction condition, and combined with the scene attributes of the virtual scene and the real-time state of the interactive scene, generate a multi-dimensional dynamic response candidate set. The response content in the dynamic response candidate set covers visual feedback, interactive feedback, and scene state feedback.

[0086] In this embodiment, the virtual scene's attributes include "track scene," "vehicle model," and "realistic visual style." The real-time state of the interactive scene includes "current vehicle speed" and "current track segment is a curve." Based on feedback logic triggering reconstruction conditions, a multi-dimensional dynamic response candidate set is generated by combining scene attributes and real-time state. Visual feedback candidates include "accelerated vehicle steering animation" and "highlighted track curve prompts"; interactive feedback candidates include "increased controller force feedback intensity" and "vibration feedback after continuous clicks"; scene state feedback candidates include "increased vehicle steering sensitivity" and "temporary adjustment of track friction." The dynamic response candidate set covers response content in these three dimensions.

[0087] Step 1651: Obtain multiple condition elements of the feedback logic triggering reconstruction condition, extract the parameter constraint range corresponding to each condition element, and establish the association mapping relationship between the condition elements and the response output.

[0088] In this embodiment, the feedback logic triggering reconstruction condition is "steering angular velocity > 25° / s and operating frequency > 2Hz", with the condition elements including "steering angular velocity > 25° / s" and "operating frequency > 2Hz". The parameter constraint ranges corresponding to each condition element are extracted: the parameter constraint range for "steering angular velocity > 25° / s" is steering angular velocity parameter > 25° / s, and the parameter constraint range for "operating frequency > 2Hz" is operating frequency parameter > 2Hz. An association mapping relationship is established between the condition elements and the response outputs. For example, "steering angular velocity > 25° / s" is associated with the response outputs of "accelerated vehicle steering animation" and "increased handle force feedback intensity", and "operating frequency > 2Hz" is associated with the response outputs of "continuous click vibration feedback" and "increased vehicle steering sensitivity".

[0089] Step 1652: Collect scene attribute data of the virtual scene, classify and organize it according to scene environment parameters, interactive object attributes, and scene physical rules, extract key indicators that affect the presentation of interactive feedback from each type of attribute, and generate scene attribute feature set.

[0090] In this embodiment, scene attribute data of the virtual scene is collected. Scene environment parameters include "track type is a curve", "light intensity is medium", and "scene complexity is high"; interactive object attributes include "vehicle model is a sports car", "vehicle current speed is high", and "controller type is a VR positioning controller"; scene physics rules include "vehicle steering physics engine parameters", "track friction parameters", and "collision detection rules". After being sorted by category, key indicators affecting the interactive feedback presentation are extracted, such as "track type" and "light intensity" in the scene environment parameters, "vehicle model" and "vehicle current speed" in the interactive object attributes, and "vehicle steering physics engine parameters" and "track friction parameters" in the scene physics rules, to generate a scene attribute feature set.

[0091] Step 1653: Match the element parameters of the feedback logic-triggered reconstruction condition with the scene attribute feature set, and filter out the associated features related to condition activation in the scene attributes; based on the dependency relationship between the associated features and the condition elements and the results of historical data analysis, calculate the weight quantification value of the influence of each associated feature on condition execution.

[0092] In this embodiment, the element parameters for triggering the reconstruction condition by the feedback logic are "steering angle velocity > 25° / s" and "operation frequency > 2Hz". These are matched with the scene attribute feature set to filter out the associated features related to condition activation in the scene attributes, such as "track type is a curve", "vehicle current speed is high", and "vehicle steering physics engine parameters". Based on the dependency relationship between the associated features and the condition elements, such as "track type is a curve" depending on the activation of the "steering angle velocity > 25° / s" condition, and "vehicle current speed is high" depending on the activation of the "operation frequency > 2Hz" condition, combined with the results of historical data analysis, the influence weight quantification value of each associated feature on the condition execution is calculated. The influence weight of "track type is a curve" is 0.7, the influence weight of "vehicle current speed is high" is 0.6, and the influence weight of "vehicle steering physics engine parameters" is 0.5.

[0093] Step 1654: Combine the real-time status data of the interaction scenario and correct the influence weight quantification value of the associated features according to the preset weight correction rules; based on the corrected weight quantification value, determine the core associated nodes of the scenario attributes and feedback logic.

[0094] In this embodiment, the real-time status data of the interactive scene includes "vehicle current speed is extremely high" and "the current track segment is a sharp curve". The preset weight correction rule is "if the vehicle current speed is extremely high, increase the influence weight of the 'vehicle current speed is high' associated feature by 0.1; if the current track segment is a sharp curve, increase the influence weight of the 'track type is curve' associated feature by 0.1". Based on this rule, the influence weight quantification value of the associated features is corrected. After correction, the influence weight of "track type is curve" is 0.8, the influence weight of "vehicle current speed is high" is 0.7, and the influence weight of "vehicle steering physics engine parameters" remains at 0.5. Based on the corrected weight quantification value, the core associated nodes of scene attributes and feedback logic are determined to be "track type is curve" and "vehicle current speed is high".

[0095] Step 1655: For the visual feedback dimension, based on the visual style parameters in the scene attributes and the display requirements in the real-time state, generate visual feedback candidates with different effect levels; for the interactive feedback dimension, based on the timing requirements of the feedback logic triggering reconstruction conditions and the user interaction rhythm in the real-time state, design interactive feedback candidates that match the action execution timing; for the scene state feedback dimension, based on the state adjustment rules in the scene attributes and the current scene state in the real-time state, generate state feedback candidates that conform to the scene development logic.

[0096] In this embodiment, the visual style parameter is "realistic style," and the real-time display requirement is "highlighting the curve prompt." Visual feedback candidates are generated, such as low-level "slightly highlighted curve prompt," medium-level "medium highlighted curve prompt with a turn arrow," and high-level "high highlighted curve prompt with accelerated vehicle steering animation." The timing requirement for the feedback logic triggering reconstruction condition is "feedback must respond within 50ms after the action is triggered." The user interaction rhythm in the real-time state is "operation frequency 3Hz," and interactive feedback candidates are designed, such as "handle force feedback intensity increases with steering angular velocity" and "continuous click vibration feedback matches a 3Hz operation frequency." The state adjustment rule in the scene attributes is "increased vehicle steering sensitivity at curves," and the current scene state in the real-time state is "vehicle current speed is extremely high." State feedback candidates are generated, such as "vehicle steering sensitivity increased by 20%" and "track friction temporarily reduced by 10%."

[0097] Step 1656: Generate a multi-dimensional dynamic response candidate set based on the candidate options in terms of visual feedback dimension, interactive feedback dimension, and scene state feedback dimension.

[0098] In this embodiment, candidate options for visual feedback include "slightly highlighted cornering prompts", "medium highlighted cornering prompts with displayed turn arrows", and "highlighted cornering prompts with accelerated vehicle steering animation"; candidate options for interactive feedback include "increased handle force feedback intensity with steering angular velocity" and "vibration feedback matching a 3Hz operation frequency"; candidate options for scene state feedback include "20% increase in vehicle steering sensitivity" and "10% temporary reduction in track friction". These candidate options are combined to generate a multi-dimensional dynamic response candidate set, where each candidate solution includes one or more combinations of visual, interactive, and scene state feedback.

[0099] Step 166: Input the interaction intent description and the dynamic response candidate set into a preset intent response matching model. Calculate the semantic association metric value between the interaction intent and each dynamic response candidate scheme through semantic association analysis. Calculate the adaptation metric value between each dynamic response candidate scheme and the current virtual scene through scene adaptability evaluation. Based on a preset matching strategy, combine the semantic association metric value and the adaptation metric value to select the dynamic response scheme with the best overall evaluation as the optimal dynamic response scheme. Update the real-time interaction feedback of the virtual scene according to the optimal dynamic response scheme, and synchronously associate and store the response execution information with the corresponding time sequence node markers.

[0100] In this embodiment, the preset intent-response matching model is a two-layer neural network. The first layer calculates semantic relevance, and the second layer calculates scene adaptability. The interaction intent description and the dynamic response candidate set are input into the model. Semantic relevance metric values ​​between the interaction intent and each candidate solution are calculated through semantic relevance analysis. For example, the semantic relevance of the candidate solution "highlighting of curve prompts and acceleration of vehicle steering animation + increase in handle force feedback intensity with steering angular velocity + 20% increase in vehicle steering sensitivity" is 0.9. Scene adaptability metric values ​​between each candidate solution and the current virtual scene are calculated through scene adaptability evaluation; this candidate solution has an adaptability of 0.85. The preset matching strategy is: comprehensive evaluation = semantic relevance × 0.6 + adaptability × 0.4. The comprehensive evaluation of this candidate solution is calculated as 0.9 × 0.6 + 0.85 × 0.4 = 0.88, which is selected as the optimal dynamic response solution. Real-time interaction feedback is updated according to this solution, and the response execution information is synchronously associated and stored with the corresponding time-series node markers.

[0101] In some embodiments, it also includes: Step 170: Extract the semantic information of the game system and the interaction adaptation data of the real-time dynamic response, and locate the associated deviation nodes in the semantic encoding process by combining the semantic mapping record of the effective behavioral features.

[0102] In this embodiment, the interaction adaptation data includes users' subjective ratings and objective indicators of real-time dynamic responses, such as response latency and feedback accuracy. The semantic mapping record stores the mapping chain from effective behavioral features to semantic elements. By comparing the interaction adaptation data with the expected effect, the semantic elements corresponding to interactions with low ratings or high latency are located, and the feature mapping node that generated the element is traced back to the node associated with the deviation.

[0103] Step 171: Import the associated deviation nodes, corresponding effective behavioral features, and semantic element set into the semantic coding optimization model. With the goal of increasing the accuracy of the transmission of interactive intent description, adjust the matching weights of features and semantic factors in the semantic mapping rules to generate coding logic correction parameters.

[0104] In this embodiment, the semantic encoding optimization model is an online learning model. It takes as input the associated deviation node, the corresponding effective behavioral features and the set of semantic elements, calculates the gradient of the matching weights of features and semantic factors in the current semantic mapping rule on the accuracy of the transmission of the interactive intent description, updates the weights using the stochastic gradient descent algorithm, reduces the transmission accuracy loss function, and generates encoding logic correction parameters, such as adjusting the matching weights of the "steering angular velocity" feature and the "rapid steering" semantic factor.

[0105] Step 172: Update the feature parsing task of semantic encoding processing based on the modified parameters of the encoding logic, and filter out candidate semantic factors with semantic conflicts through the newly added semantic factor correlation verification task.

[0106] In this embodiment, the feature parsing task of semantic coding processing is updated according to the coding logic correction parameters, such as adjusting the feature extraction threshold and optimizing the feature matching algorithm. A new semantic factor correlation verification task is added, which stores the compatibility / conflict relationships between semantic factors based on the knowledge graph, such as the conflict between "turning" and "stationary". After the candidate semantic factors are generated, the conflict relationship between factors is checked, and candidate semantic factors that conflict with the selected factors are filtered out.

[0107] Step 173: Re-encode the effective behavioral features using the updated encoding process to generate an optimized semantic element set. Compare the optimized semantic element set with the original semantic element set to extract the optimization effect features of the encoding logic.

[0108] In this embodiment, the effective behavioral features are re-encoded using an updated encoding process to generate an optimized set of semantic elements. The optimized set of semantic elements is then compared with the original set of semantic elements to calculate the similarity of intent descriptions (such as BLEU scores), semantic element completeness, number of conflict factors, etc., and to extract the optimization effect features of the encoding logic, such as "BLEU score increased by 0.15" and "number of conflict factors decreased by 2".

[0109] Step 174: Based on the optimization effect features, update the parameter configuration of the semantic mapping rule, and apply the updated encoding logic to the semantic processing of the next non-preset interactive action data.

[0110] In this embodiment, based on the optimization effect characteristics "BLEU score increased by 0.15" and "number of conflict factors decreased by 2", the encoding logic optimization is determined to be effective. The parameter configuration of the semantic mapping rule is then updated, such as adjusting the matching weights between features and semantic factors and updating the semantic factor conflict relationship database. The updated encoding logic is then applied to the semantic processing of the next non-preset interactive action data to improve the accuracy of semantic encoding.

[0111] In some embodiments, it also includes: Step 180: Collect multiple sets of historical non-preset interactive action data and corresponding behavioral feature sets as training sample sets, perform temporal node alignment and feature dimension annotation on the non-preset interactive action data in the training sample set, and generate standardized training samples.

[0112] In this embodiment, historical non-preset interactive action data is collected, including action data sequences and behavioral feature sets, and the ground truth values ​​of behavioral features are labeled manually or semi-automatically. Temporal dynamic time warping is used to align the start and end points of actions in different samples, and the feature dimension is unified as a fixed-length vector to generate standardized training samples.

[0113] Step 181: Construct a deep learning model for behavioral feature extraction, divide the standardized training samples into a training subset and a validation subset according to a preset ratio, and initialize the feature extraction layer parameters of the deep learning model.

[0114] In this embodiment, the deep learning model adopts an encoder-decoder structure, where the encoder is a temporal convolutional network layer and the decoder is a fully connected layer. Standardized training samples are divided into a training subset and a validation subset in a 7:3 ratio, and the parameters of the feature extraction layer, such as the kernel weights and bias parameters of the temporal convolutional network layer, are initialized.

[0115] Step 182: Input the non-preset interactive action data in the training subset into the deep learning model, extract the temporal correlation features of the actions through the temporal convolutional layer, extract the continuity features of the spatial coordinate trajectory through the spatial attention layer, and fuse the temporal correlation features and the continuity features to generate a set of predicted behavior features.

[0116] In this embodiment, non-preset interactive action data from the training subset is input into the deep learning model. The temporal convolutional layer uses dilated causal convolution to extract temporal correlation features, and the spatial attention layer calculates attention weights for the key point coordinate sequence to highlight key joints. The temporal correlation features and continuous features are fused and the decoder outputs a set of predicted behavior features.

[0117] Step 183: Calculate the feature deviation between the predicted behavior feature set and the sample labeled behavior feature set, and adjust the convolution kernel parameters and attention weights of the deep learning model based on the feature deviation using the backpropagation algorithm.

[0118] In this embodiment, a smooth L1 loss is used to calculate the feature deviation between the predicted behavior feature set and the sample labeled behavior feature set. The gradient of the loss function with respect to the model parameters is calculated through the backpropagation algorithm, and the convolution kernel parameters of the temporal convolutional layer and the attention weights of the spatial attention layer are adjusted.

[0119] Step 184: Import the validation subset to verify the feature extraction effect of the adjusted deep learning model. Calculate the feature matching metric and extraction efficiency index, and conduct a comprehensive evaluation based on the preset evaluation criteria. Select the optimal parameter configuration of the model, apply the optimal parameter configuration to the model for solidification, and obtain the trained deep learning model.

[0120] In this embodiment, a validation subset is imported to validate the adjusted deep learning model, calculating the feature matching degree (mean cosine similarity between the prediction and the ground truth) and extraction efficiency (single sample processing time). The preset evaluation criteria are feature matching degree > 0.85 and extraction efficiency < 10ms. The optimal parameter configuration of the model that meets the criteria is selected, and it is fixed and saved as a model file to obtain the trained deep learning model.

[0121] In some embodiments, it also includes: Step 190: Summarize the standardized behavioral feature subset, intent confidence set, and interaction intent description to generate a model training dataset. Add intent labels to the standardized behavioral feature subset in the model training dataset to generate labeled training samples.

[0122] In this embodiment, standardized behavioral feature subsets, intent confidence sets, and interaction intent descriptions generated during historical processing are summarized. Intent labels are added to each standardized behavioral feature subset. The labels are derived from manual review or the final game interaction results, and labeled training samples are generated.

[0123] Step 191: Call the initially constructed interaction intent probability model, input the labeled training samples into the initially constructed interaction intent probability model, and obtain the output local intent confidence and correlation analysis results.

[0124] In this embodiment, the initially constructed interaction intent probability model uses randomly initialized parameters, inputs labeled training samples into the model, and obtains the local intent confidence and association analysis results output by the temporal association processing branch, the pattern matching processing branch, and the association anomaly detection processing branch.

[0125] Step 192: Calculate the prediction bias of the confidence of each local intent based on the true intent corresponding to the label, and construct a multi-dimensional loss function. The multi-dimensional loss function covers temporal correlation bias, pattern matching bias and anomaly detection bias.

[0126] In this embodiment, the prediction bias of the confidence of each local intent is calculated based on the true intent corresponding to the label. The temporal correlation bias is calculated using KL divergence, the pattern matching bias is calculated using cross-entropy, and the anomaly detection bias is calculated using the inverse of F1 score. A multi-dimensional loss function is constructed with a weight ratio of 4:3:3.

[0127] Step 193: Based on the calculation results of the multi-dimensional loss function, the gradient descent method is used to sequentially adjust the sliding window parameters of the temporal association processing branch, the similarity weighting coefficient of the pattern matching processing branch, and the baseline range parameters of the association anomaly detection processing branch in the interaction intent probability model.

[0128] In this embodiment, the Adam optimizer is used to calculate the gradient of the loss function with respect to the model parameters, and the sliding window parameters of the temporal correlation processing branch, the similarity weighting coefficient of the pattern matching processing branch, and the baseline range parameters of the correlation anomaly detection processing branch are adjusted in sequence.

[0129] Step 194: Iteratively execute the sample input, bias calculation and parameter adjustment process until the matching accuracy between the model output intent confidence set and the real intent label reaches the preset accuracy value, and obtain the target interaction intent probability model.

[0130] In this embodiment, the sample input, bias calculation and parameter adjustment process is executed iteratively. An early stop mechanism is set. If the validation set loss does not decrease for 5 consecutive rounds, the process stops until the matching accuracy between the model output intent confidence set and the real intent label reaches a preset accuracy value of 0.92, thus obtaining the target interaction intent probability model.

[0131] Step 195: Introduce newly collected non-preset interactive action data samples to perform a generalization test on the target interactive intent probability model, and fine-tune the output fusion strategy of the target interactive intent probability model based on the test results.

[0132] In this embodiment, newly collected non-preset interactive action data samples are introduced to perform generalization tests on the target interactive intent probability model. The intent recognition accuracy of the test samples is calculated. If the accuracy drops by more than 5%, the output fusion strategy of the model is fine-tuned, such as adjusting the weighted fusion weights of the confidence of the three branches, and using grid search on new data to redetermine the optimal weight combination.

[0133] This application's embodiments achieve accurate intent understanding and dynamic scene adaptation for non-preset interactive actions in virtual reality interactive games. First, by capturing non-preset interactive action data containing temporal node markers in real time, and combining temporal correlation and spatial continuity to extract action trend features, operation rhythm features, and intensity fluctuation features, it breaks through the definition boundaries of traditional preset action libraries for interactive behaviors, covering impromptu actions generated by users due to unexpected scene situations. Second, it uses an interaction intent probability model to perform confidence screening on the behavioral feature set. Through multi-branch collaborative calculation of intent confidence sets using temporal correlation, pattern matching, and anomaly detection, it effectively filters invalid action noise, improving the system's accuracy in recognizing the user's true interaction intent. Furthermore, through semantic mapping rules, it transforms effective behavioral features into game system semantic information containing interactive intent descriptions, establishing a mapping bridge from action features to semantic intent, and solving the semantic gap problem between unstructured action data and structured scene rules. Finally, based on the virtual scene type, adaptation rules are invoked, matching nodes are determined through logical matching, and parameter distribution is adjusted to generate a dynamic adaptation scheme. The interaction intent description is combined with the updated feedback logic trigger conditions and real-time dynamic response. The response time is associated with the time sequence node mark, which ensures the real-time performance and scene adaptability of the interaction feedback, enhances the immersion and personalization of virtual scene interaction, and provides richer interaction methods and a more intelligent feedback mechanism for virtual reality interactive games.

[0134] Please see Figure 2 The figure is a schematic diagram of the basic structure of an interactive game data processing system 200 provided in an embodiment of this application. The interactive game data processing system 200 includes: a processor 201; a storage device 202 on which a computer program 2020 is stored; and a network interface 203 for providing network communication functions. When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the interactive game data processing methods for virtual reality applications described above.

[0135] Please see Figure 3 This application provides a functional block diagram of an interactive game data processing device, which includes: The motion data capture module is used to capture in real time non-preset interactive motion data generated by users in a virtual scene through VR interactive devices. The non-preset interactive motion data includes time sequence node markers for motion execution. The behavior feature extraction module is used to extract behavior features from the non-preset interactive action data by combining the temporal correlation of action execution and the continuity of spatial coordinate trajectory, and generate a set of behavior features including action trend features, operation rhythm features and force fluctuation features. The behavior feature filtering module is used to input the behavior feature set into a pre-built interaction intent probability model, and use the intent confidence set output by the interaction intent probability model to filter out effective behavior features from the behavior feature set; The semantic encoding processing module is used to perform semantic encoding processing on the effective behavioral features according to the semantic mapping rules preset by the game system to obtain a set of semantic elements, and generate game system semantic information containing interactive intent descriptions through the association and combination of the set of semantic elements. The scene dynamic adaptation module is used to call the matching scene adaptation rules according to the type of the virtual scene, logically match the semantic information of the game system and the scene adaptation rules to determine the matching node between the semantic information of the game system and the interaction requirements of the virtual scene, adjust the parameter distribution of the scene adaptation rules based on the matching node, and generate a scene dynamic adaptation scheme. The dynamic response update module is used to adjust the feedback logic triggering conditions of the virtual scene based on the scene dynamic adaptation scheme and the interaction intent description, and update the real-time dynamic response of the virtual scene to the non-preset interaction action according to the feedback logic triggering conditions. The interaction response timeliness of the real-time dynamic response is associated with the corresponding time sequence node marker.

[0136] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0137] Furthermore, it should be noted that this application also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of the interactive game data processing system reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the interactive game data processing system to perform the aforementioned... Figure 1 The methods described in the corresponding embodiments are already known, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program product embodiments related to this application, please refer to the description of the method embodiments of this application.

[0138] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. A method for processing interactive game data using virtual reality, characterized in that, include: Real-time capture of non-preset interactive action data generated by users in a virtual scene through VR interactive devices, wherein the non-preset interactive action data includes time sequence node markers of action execution; By combining the temporal correlation of action execution with the continuity of spatial coordinate trajectory, behavioral features are extracted from the non-preset interactive action data to generate a set of behavioral features including action trend features, operation rhythm features, and intensity fluctuation features. The behavioral feature set is input into a pre-built interaction intent probability model, and the effective behavioral features are selected from the behavioral feature set using the intent confidence set output by the interaction intent probability model. The effective behavioral features are semantically encoded according to the semantic mapping rules preset by the game system to obtain a set of semantic elements. The semantic elements are then combined to generate semantic information of the game system that contains a description of the interactive intent. According to the type of the virtual scene, the matching scene adaptation rules are invoked. The semantic information of the game system and the scene adaptation rules are logically matched to determine the matching nodes between the semantic information of the game system and the interaction requirements of the virtual scene. Based on the matching nodes, the parameter distribution of the scene adaptation rules is adjusted and a dynamic scene adaptation scheme is generated. Based on the scene dynamic adaptation scheme and the interaction intent description, the feedback logic triggering conditions of the virtual scene are adjusted, and the real-time dynamic response of the virtual scene to the non-preset interaction action is updated according to the feedback logic triggering conditions. The interaction response timeliness of the real-time dynamic response is associated with the corresponding time sequence node marker.

2. The method as described in claim 1, characterized in that, The step of inputting the set of behavioral features into a pre-constructed interaction intent probability model, and using the intent confidence set output by the interaction intent probability model to filter out effective behavioral features from the set of behavioral features includes: Based on the constituent dimensions of the behavioral feature set, extract the common temporal node markers of each feature in the behavioral feature set as association anchors, perform feature dimension unification for the attribute differences of each feature, and generate a standardized behavioral feature subset containing temporal anchors. The action trend features and corresponding temporal anchors are imported into the preset temporal association processing branch in the interaction intent probability model. The temporal dependencies of features at different time granularities are extracted through a variable step-size sliding window, and a directed action trend graph is constructed. The occurrence probability of each transition path in the directed action trend graph is calculated and mapped to the first intent association coefficient in combination with the transition time. Based on the coefficient distribution and the temporal continuity of the temporal anchors, the first local intent confidence corresponding to the action trend features is generated, and the trend status label corresponding to each temporal anchor is output synchronously. The pattern matching processing branch of the interaction intent probability model receives the temporal anchor points and trend status labels output by the temporal association processing branch. The operation rhythm feature is split into segmented rhythm features according to the temporal anchor points. Periodic splitting is performed on each segmented rhythm feature and it is compared with the preset interaction rhythm template library to obtain the segment similarity. The weighted average of the segment similarities is converted into a second intent association coefficient. The second local intent confidence corresponding to the operation rhythm feature is generated by combining the intent labels of the preset interaction rhythm template library and the association anchor points. The matching fit between the segmented rhythm features and the trend status labels is output synchronously. Activate the correlation anomaly detection processing branch of the interaction intent probability model, import the intensity fluctuation feature, time sequence anchor point, trend state label and segmented rhythm feature, determine the intensity benchmark range of different time sequence stages based on the trend state label, adjust the dynamic floating threshold of the benchmark range in combination with the periodic characteristics of the segmented rhythm feature, identify the fluctuation segment in the intensity fluctuation feature that exceeds the dynamic benchmark range, quantify the effective interaction intensity change judgment result of the fluctuation segment and the synchronization coefficient of the segmented rhythm feature into a third intent correlation coefficient, generate the third local intent confidence corresponding to the intensity fluctuation feature, and output the correlation analysis result of the fluctuation segment with the operation rhythm feature and the action trend feature; Using the associated anchor point as an index, the first local intent confidence, the second local intent confidence, and the third local intent confidence are standardized. The standardized local intent confidences are then weighted and fused according to preset weights to obtain a fused confidence. The trend status label, the matching fit, and the association analysis result are quantified into numerical features. Based on preset association rules, the numerical features are combined to generate a quantified feature association matrix. According to the quantified values ​​of each element in the feature association matrix and the fused confidence, the comprehensive intent confidence corresponding to each feature subset within the standardized behavioral feature subset is calculated to form an intent confidence set. The feature subsets in the intent confidence set that exceed the target intent confidence level of the preset confidence set are taken as the effective behavioral features.

3. The method as described in claim 1, characterized in that, The process of performing semantic encoding on the effective behavioral features according to the semantic mapping rules preset by the game system to obtain a set of semantic elements, and generating game system semantic information containing interactive intent descriptions through the association and combination of the set of semantic elements, includes: Extract key feature items with intent orientation from each feature of the effective behavioral features, match the key feature items with the semantic mapping rules preset by the game system, and obtain the candidate semantic factors corresponding to each key feature item; Based on the temporal anchor points of each feature in the effective behavioral features, the candidate semantic factors are temporally sorted to obtain a candidate semantic factor sequence; the semantic association strength quantification value between adjacent semantic factors in the candidate semantic factor sequence is calculated, and isolated candidate semantic factors with semantic association strength quantification values ​​lower than the association strength threshold are eliminated according to the preset association strength threshold to obtain core semantic factors with continuous semantic association. Semantic attribute enhancement processing is performed on the core semantic factors. Semantic modification attributes are added to each core semantic factor in combination with the feature attributes of the effective behavioral features. The core semantic factors are associated with the corresponding effective behavioral features through the semantic modification attributes to generate semantic elements with attribute identifiers. The semantic elements with attribute identifiers are used to construct a semantic element network according to temporal anchor points and semantic association strength. Potential associations between semantic elements are mined in the semantic element network. Semantic connection links are supplemented based on potential associations to obtain a set of semantic elements with logical associations. Based on the semantic modification attributes and semantic associations of each semantic element in the set of semantic elements, the semantic elements are associated and combined to generate an interaction intent description that represents the user's interaction purpose. The interaction intent description is then integrated with the associated set of semantic elements to generate semantic information for the game system.

4. The method as described in claim 1, characterized in that, The step of invoking matching scene adaptation rules according to the type of the virtual scene, logically matching the game system semantic information and the scene adaptation rules to determine the matching nodes between the game system semantic information and the interaction requirements of the virtual scene, adjusting the parameter distribution of the scene adaptation rules based on the matching nodes, and generating a dynamic scene adaptation scheme includes: Extract the type features of the virtual scene, match the type features with a preset scene rule library, and call the scene adaptation rules that are compatible with the type of the virtual scene; The interaction intent description is converted into a standardized intent expression vector, the core semantic attributes related to scene interaction are extracted from the semantic element set, and an intent semantic association model is established. The intent semantic association model reflects the correspondence between interaction intent and semantic elements. The intent semantic association model is logically matched with the scene interaction requirement description in the scene adaptation rules. The semantic similarity comparison method is used to calculate the semantic similarity quantification value between the interaction intent and the scene interaction requirement. Based on the preset similarity threshold, the interaction intent with a semantic similarity quantification value exceeding the threshold and satisfying the logical constraints of the scene interaction requirement is identified. The logical position of the interaction intent that meets the conditions in the intent semantic association model and the scene adaptation rules is marked as the initial matching node. The rule execution logic of the scene adaptation rule is combined to determine whether the semantic element corresponding to the initial matching node triggers the rule execution process. The matching node is determined according to the trigger judgment result. Based on the matching node, the parameter items associated with the matching node are identified in combination with the parameter distribution of the scene adaptation rule. The value distribution of the parameter items is adjusted according to the intention association weight of the semantic element in the semantic information of the game system. Based on the adjusted parameter distribution, the matching nodes, and the scene adaptation rules, a scene dynamic adaptation scheme is generated.

5. The method as described in claim 4, characterized in that, The step of identifying parameter items associated with the matching node based on the matching node and the parameter distribution of the scene adaptation rules, and adjusting the value distribution of the parameter items according to the intent association weight of semantic elements in the semantic information of the game system, includes: Using parameter items as nodes and the mutual influence relationships between parameter items as edges, locate the core rule entry corresponding to each matching node in the parameter association network of the scene adaptation rule, track all target parameter items involved in the core rule entry, and mark the target parameter items as candidate association parameters; Based on the quantified value of the degree of influence of parameters on the rule execution effect and the quantified value of the matching between parameters and semantic elements, according to the preset association strength calculation rules, the quantified value of the association strength between each candidate association parameter and the matching node is calculated; the parameter items with the quantified value of association strength exceeding the preset strength threshold are selected as key parameter items associated with the matching node, and a correspondence table between key parameter items and matching nodes is established. The intent association weights of each semantic element in the semantic information of the game system are extracted, and the intent association weights are normalized to obtain standardized weight values. The standardized weight values ​​are matched with the key parameter items in the corresponding relationship table to determine the weight reference value corresponding to each key parameter item. The weight reference value reflects the degree of guidance of the interaction intent on parameter adjustment. Obtain the current value and adjustment range of each key parameter item. Using the weight reference value as a guide, generate a parameter adjustment strategy based on the mutual constraints between parameters. Perform multiple rounds of parameter adjustment operations according to the parameter adjustment strategy until the adjusted parameter distribution matches the requirements of the interaction intent.

6. The method as described in claim 1, characterized in that, The step of adjusting the feedback logic triggering conditions of the virtual scene based on the scene dynamic adaptation scheme and the interaction intent description, and updating the real-time dynamic response of the virtual scene to the non-preset interaction action according to the feedback logic triggering conditions, includes: Based on the description of interaction intent, the rules, constraints and parameter adjustment directions related to virtual scene feedback in the scene dynamic adaptation scheme are analyzed, and the core demands of interaction intent are transformed into a priority sequence of feedback logic adjustment, so that the optimization direction of feedback logic is consistent with the user's interaction purpose. Based on the priority sequence, extract the constituent elements of the current feedback logic triggering conditions of the virtual scene, and combine them with the adjusted parameter distribution in the scene dynamic adaptation scheme to identify the logic nodes in the current feedback logic triggering conditions that do not match the scene adaptation requirements and interaction intentions, and obtain the set of triggering condition optimization target points. The preset feedback logic reasoning model is invoked to perform logical association reasoning on the set of target points for trigger condition optimization, the description of interaction intent, and the scene dynamic adaptation parameters to generate a trigger condition adjustment strategy. The feedback logic triggering reconstruction condition of the virtual scene is determined according to the triggering condition adjustment strategy. An association mapping is established between the feedback logic triggering reconstruction condition and the time sequence node marker in the non-preset interactive action data, so that the activation timing of the feedback logic triggering reconstruction condition matches the time characteristics of the action execution. Based on the feedback logic triggering the reconstruction condition, combined with the scene attributes of the virtual scene and the real-time state of the interactive scene, a multi-dimensional dynamic response candidate set is generated. The response content in the dynamic response candidate set covers visual feedback, interactive feedback and scene state feedback. The interaction intent description and the dynamic response candidate set are input into a preset intent-response matching model. Semantic association analysis is used to calculate the semantic association metric value between the interaction intent and each dynamic response candidate scheme. Scene adaptability evaluation is used to calculate the adaptability metric value between each dynamic response candidate scheme and the current virtual scene. Based on a preset matching strategy, the dynamic response scheme with the best comprehensive evaluation is selected as the optimal dynamic response scheme by combining the semantic association metric value and the adaptability metric value. The real-time interaction feedback of the virtual scene is updated according to the optimal dynamic response scheme, and the response execution information is synchronously associated and stored with the corresponding time sequence node marker.

7. The method as described in claim 6, characterized in that, The step of invoking a preset feedback logic reasoning model to perform logical correlation reasoning on the set of target points for trigger condition optimization, the description of interaction intent, and the dynamic adaptation parameters of the scene, and generating a trigger condition adjustment strategy, including: Extract the logical attributes, parameter constraints, and association levels with historical feedback logic of each optimization target in the set of optimization targets for triggering conditions. After distinguishing core targets and related targets according to the association level, construct a target feature matrix based on target type, attribute features, and association weight. Extract the core demand items, scenario adaptation expectations, and response timeliness requirements of the interaction intent description, map the core demand items into quantitative intent indicators, transform the scenario adaptation expectations into scenario constraint parameters, and generate timeliness thresholds by associating time sequence node markers with response timeliness requirements to obtain the intent quantification feature vector. The scene dynamic adaptation parameters are classified into scene basic parameters, interaction response parameters, and state adjustment parameters. The deviation between the current value and the optimal range of each type of parameter is calculated. The parameter influence coefficient is generated by combining the influence weight of the parameter on the feedback logic, and the parameter feature set is obtained. Align the target feature matrix, the intent quantification feature vector, and the parameter feature set into the input feedback logic reasoning model, and assign the highest weight to the core target features and the core intent appeal features, and assign secondary weight to the associated target features and parameter influence coefficients. Through the feedback logic reasoning model, the relationship between features is established based on the predicate logic rules, the matching degree between the target optimization requirement features and the intention quantitative indicator features is calculated, and the feasibility boundary of target adjustment is defined by the parameter influence coefficient, generating multiple preliminary logical reasoning paths with quantitative matching degree evaluation. Using historical interaction data and feedback logic execution cases from the virtual scenario, each inference path is evaluated and filtered based on the execution success rate quantification value and the interaction adaptation quantification value. The remaining valid inference paths after filtering are deduplicated and optimized. The target adjustment direction, parameter modification range and logical association rules are extracted from the deduplicated and optimized inference paths. Based on the core target adjustment strategy and supplemented by the related target adjustment strategy, the execution order of each strategy is determined by combining the timeliness requirements in the intent quantification feature vector, and the trigger condition adjustment strategy is generated.

8. The method as described in claim 6, characterized in that, The process, based on the feedback logic triggering the reconstruction condition and combining the scene attributes of the virtual scene and the real-time state of the interactive scene, generates a multi-dimensional dynamic response candidate set, including: Obtain multiple condition elements of the feedback logic triggering the reconstruction condition, extract the parameter constraint range corresponding to each condition element, and establish the association mapping relationship between the condition elements and the response output; Collect scene attribute data of virtual scenes, classify and organize them according to scene environment parameters, interactive object attributes, and scene physical rules, extract key indicators that affect the presentation of interactive feedback from various attributes, and generate scene attribute feature sets; The element parameters of the feedback logic-triggered reconstruction conditions are matched with the scene attribute feature set to filter out the associated features related to condition activation in the scene attributes; based on the dependency relationship between the associated features and the condition elements and the results of historical data analysis, the weight quantification value of the influence of each associated feature on condition execution is calculated. By combining real-time status data of the interaction scenario and according to the preset weight correction rules, the influence weight quantification value of the associated features is corrected; based on the corrected weight quantification value, the core associated nodes of the scenario attributes and feedback logic are determined. For the visual feedback dimension, visual feedback candidates with different effect levels are generated by combining the visual style parameters in the scene attributes and the display requirements in the real-time state. For the interactive feedback dimension, interactive feedback candidates that match the action execution sequence are designed based on the timing requirements of the feedback logic trigger reconstruction conditions and the user interaction rhythm in the real-time state. For the scene state feedback dimension, state feedback candidates that conform to the scene development logic are generated by combining the state adjustment rules in the scene attributes and the current scene state in the real-time state. A multi-dimensional dynamic response candidate set is generated based on candidates from the dimensions of visual feedback, interactive feedback, and scene state feedback.

9. An interactive game data processing system, characterized in that, include: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the interactive game data processing method for virtual reality as described in any one of claims 1-8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the interactive game data processing method for virtual reality as described in any one of claims 1-8.

Images