AI-driven interactive AR content intelligent generation and distribution method and system

Abstract

The invention provides an AI-driven interactive AR content intelligent generation and distribution method and system, and relates to the technical field of AI driving, and the method comprises the steps: obtaining scene data, constructing a three-dimensional semantic map, calculating a scene region semantic entropy value, and constructing a scene semantic evolution tree; matching the candidate content with the semantic change trend, dividing content groups, and generating a trigger condition and an update strategy; collecting user interaction data to establish an interaction behavior model, and calculating the interaction responsivity of the content group; and finally, grading based on the interaction responsivity to determine the pushing priority, and constructing a directed pushing path diagram to guide the activation sequence of the content groups. Through the semantic entropy-driven content organization method and the interactive behavior-based intelligent distribution mechanism, accurate pushing and smooth transition of augmented reality content are realized, and the user interaction experience is improved.

Description

Technical Field

[0001] The present invention relates to AI-driven technologies, and particularly to an AI-driven interactive AR content intelligent generation and distribution method and system. Background Art

[0002] Augmented reality technology provides an immersive interactive experience for users by overlaying virtual content onto real-world scenes. With the development of mobile terminals and edge computing technologies, AI-based AR content generation and distribution technologies have been widely applied. Current AR systems mainly achieve the overlay display of virtual content through scene understanding and content matching, and optimize the content push strategy based on user interaction data to enhance the user experience.

[0003] However, the existing technologies still have the following deficiencies: First, traditional scene understanding methods lack in-depth analysis of the evolution rules of scene semantics, and cannot accurately grasp the trend of scene semantic changes, resulting in insufficient content matching accuracy; Second, the existing content organization methods usually adopt static partitioning strategies, without fully considering the impact of user interaction behaviors on content relevance, and it is difficult to adapt to dynamic interaction scenarios; Third, the current content distribution mechanism often adopts a simple priority strategy, lacking a systematic modeling of the semantic associations between content groups, and prone to causing discontinuous content switching.

[0004] The present invention aims to solve the above technical problems, and proposes a scene understanding method based on semantic entropy value and an interaction-driven content distribution mechanism. By constructing a scene semantic evolution tree, it can accurately grasp the trend of scene changes, optimize the content organization strategy based on the interaction behavior model, and combine the directed push path of semantic associations to achieve smooth content transition, thereby enhancing the interaction experience of the AR system. Summary of the Invention

[0005] Embodiments of the present invention provide an AI-driven interactive AR content intelligent generation and distribution method and system, which can solve the problems in the existing technologies.

[0006] In the first aspect of the embodiments of the present invention, An AI-driven interactive AR content intelligent generation and distribution method is provided, including: Obtaining scene data collected by a user terminal, and constructing a three-dimensional semantic map of a target scene based on the scene data; calculating the semantic entropy value of each region in the scene according to the three-dimensional semantic map, constructing a scene semantic evolution tree based on the semantic entropy value, and obtaining the semantic change trend of the target scene by updating the node states and transition weights in the scene semantic evolution tree in real time; Calculate the matching degree between the candidate content in the augmented reality content library and the semantic change trend, divide the relevant candidate content into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; Collect the interaction data between the user and the content group, where the interaction data includes interaction type, interaction duration, interaction result, and scene state change. Establish an interaction behavior model based on the interaction data, and calculate the interaction response degree of each content group through the interaction behavior model; Grade the content groups according to the interaction response degree, determine the push priority based on the grading result, and construct a directed push path graph in combination with the semantic association degree between the content groups. The directed push path graph is used to guide the activation order of the content groups, and realize the smooth transition of the augmented reality experience through the ordered activation of the content groups.

[0007] Calculate the semantic entropy value of each region in the scene according to the three-dimensional semantic map, construct a scene semantic evolution tree based on the semantic entropy value, and obtain the semantic change trend of the target scene by updating the node state and transition weight in the scene semantic evolution tree in real time, including: Divide the scene space of the three-dimensional semantic map into voxel grids, calculate the probability distribution of the appearance of each semantic category data in each voxel grid, and calculate the semantic entropy value of each voxel grid based on the probability distribution. The semantic entropy value is used to characterize the semantic complexity of the voxel grid; Obtain the neighborhood voxel set of each voxel grid, calculate the spatial distance between each voxel in the neighborhood voxel set and the target voxel, determine the spatial association weight based on the spatial distance, and perform weighted calculation on the semantic entropy value according to the spatial association weight to obtain the weighted semantic entropy value considering neighborhood influence; Construct a scene semantic evolution tree based on the weighted semantic entropy value, calculate the semantic state change amount, state transition probability, and state transition time interval between adjacent nodes in the scene semantic evolution tree, and use the semantic state change amount, the state transition probability, and the state transition time interval as the attributes of the connection edges between the nodes in the scene semantic evolution tree; Calculate the similarity between the nodes in the scene semantic evolution tree. The similarity is obtained through the weighted combination of the cosine similarity of the semantic state vector, the Euclidean distance of the entropy value distribution matrix, and the time decay function. Dynamically update the scene semantic evolution tree based on the similarity to obtain the semantic change trend of the scene.

[0008] Construct a scene semantic evolution tree based on the weighted semantic entropy value, and calculate the semantic state change amount, state transition probability, and state transition time interval between adjacent nodes in the scene semantic evolution tree, including: Extract the semantic attribute set of the scenario based on the weighted semantic entropy value. The semantic attribute set includes object category, spatial location, object attributes, and relationships between objects. By performing threshold segmentation on the weighted semantic entropy value and extracting the semantic region, obtain the semantic feature vector for each semantic region, and normalize the semantic feature vector to obtain the attribute intensity value corresponding to each semantic attribute; Construct the node representation of the scenario semantic evolution tree according to the semantic attribute set and the attribute intensity value. Each node contains a state vector and an entropy value distribution matrix. Construct the state vector by organizing the attribute intensity values into a vector form according to a preset attribute sorting rule, and construct the entropy value distribution matrix by organizing the weighted semantic entropy values into a matrix form according to the spatial position relationship; Identify adjacent node pairs in the semantic evolution tree. For each adjacent node pair, calculate the difference between the components of the state vectors between the nodes, divide the difference by the maximum value of the corresponding component and introduce a smoothing factor to obtain the normalized semantic state change amount; Calculate the two-norm of the semantic state change amount to obtain the state distance, perform an exponential operation on the negative of the state distance to obtain the initial probability value, and normalize the initial probability values of all adjacent nodes of each node to obtain the state transition base probability; Calculate the cosine similarity of the corresponding semantic attributes in the state vector to obtain the similarity value of each attribute, and perform a weighted sum of the similarity value and a preset importance weight to obtain the state transition time interval.

[0009] Calculate the matching degree between the candidate content in the augmented reality content library and the semantic change trend. Divide the candidate content with relevance into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map, including: Collect the feature information of the candidate content in the augmented reality content library, and convert the feature information into a feature vector. The feature vector includes a visual feature component, a semantic feature component, and an interaction feature component; Perform a matching calculation between the feature vector of the candidate content and the semantic change trend. By calculating the degree of consistency between the change direction of the feature vector of the candidate content in the spatial dimension, attribute dimension, and interaction dimension and the semantic change trend, obtain the matching degree score of each candidate content; Use the vector inner product of the matching degree scores as the similarity value between the candidate contents to construct a similarity matrix; calculate the mean and standard deviation of the similarity matrix, set a dynamic clustering threshold based on the mean and the standard deviation, and divide the candidate contents with similarity higher than the dynamic clustering threshold into the same content group; Extract scene features from the three-dimensional semantic map. The scene features include spatial topological structure, semantic attribute distribution, and interaction hotspot distribution, and generate corresponding trigger conditions for each content group based on the scene features; Monitor the scene changes in the three-dimensional semantic map. When the scene changes meet the trigger conditions of the content group, calculate the update metrics for the corresponding content group. The update metrics combine the access frequency, timeliness, and scene relevance of the content group; Determine the update priority of the content group according to the update metrics. When the update priority exceeds the preset priority threshold, select the content with the highest matching degree with the current semantic change trend from the candidate content library as the update strategy.

[0010] Build an interaction behavior model based on the interaction data. Calculating the interaction response degree of each content group through the interaction behavior model includes: Extract the interaction time series features, interaction space features, and interaction modality features in the interaction data, and fuse the interaction time series features, the interaction space features, and the interaction modality features to construct a behavior feature vector; Build an interaction behavior model based on the behavior feature vector. The interaction behavior model adopts a hybrid structure of a recurrent neural network and an attention mechanism. Capture the temporal dependence relationship of the interaction behavior model through the recurrent neural network, and use the attention mechanism to highlight the importance of interaction features to generate a dynamic representation of the interaction behavior model; Calculate the interaction response degree of the content group based on the interaction behavior model. Perform similarity matching between the feature vector of the content group and the dynamic representation, and dynamically weight the result of the similarity matching through a temporal smoothing factor to obtain a response score representing the matching degree between the content group and the interaction behavior.

[0011] Construct a directed push path graph by combining the semantic association degree between content groups. The directed push path graph is used to guide the activation order of content groups, and realize a smooth transition of the augmented reality experience through the ordered activation of content groups, including: Normalize the concept weights, entity vectors, and context features in the content group to obtain a standardized feature vector, and construct a semantic representation of the content group based on the standardized feature vector; Calculate the centrality, density, and betweenness centrality of the content group, and obtain the node importance according to the weighted combination of the centrality, the density, and the betweenness centrality; construct a directed push path graph based on the node importance, and the directed push path graph is used to represent the activation dependence relationship between content groups; In the directed push path graph, a conditional probability function considering historical states is used to calculate the activation path score. The conditional probability function takes the semantic association strength, edge weight, and resource consumption of adjacent content groups as input parameters, and selects the path with the highest score as the optimal activation path; According to the access order of the optimal activation path, preload the content groups to be activated into the corresponding cache levels, calculate the content difference degree between adjacent content groups based on the semantic representation of the preloaded content groups, and calculate the context relevance in combination with real-time scenario information; Input the content difference degree and the context relevance into an adaptive state smoothing function to dynamically generate the transition time between adjacent content groups, and control the activation process of the content groups according to the transition time to achieve smooth state transition.

[0012] According to the access order of the optimal activation path, preload the content groups to be activated into the corresponding cache levels, calculate the content difference degree between adjacent content groups based on the semantic representation of the preloaded content groups, and calculate the context relevance in combination with real-time scenario information, including: Construct a time series feature vector of the content group, which includes access frequency features, time interval features, and duration features, and generate a dynamic access pattern of the content group based on the time series feature vector; Calculate the preloading priority score according to the dynamic access pattern. The preloading priority score is obtained by a weighted combination of the access frequency score, time urgency, and resource demand degree, and determine the loading order of the content groups based on the preloading priority score; Perform cache level allocation on the content groups whose loading order is determined. Determine the optimal cache level by calculating the difference between the utility value and the storage cost of the content groups in each cache level, and load the content groups into the corresponding cache levels to generate a cache allocation scheme; Obtain the multi-modal features of the content groups that have completed cache allocation, perform weighted fusion on the multi-modal features to obtain a fusion feature matrix, and generate a semantic embedding vector based on the fusion feature matrix through a non-linear mapping function; Collect the scenario state information, user behavior information, and environmental information corresponding to the semantic embedding vector, input the scenario state information, the user behavior information, and the environmental information into a context encoding function to generate a context feature representation, and calculate the context relevance between content groups according to the context feature representation.

[0013] In the second aspect of the embodiments of the present invention, Provide an AI-driven interactive AR content intelligent generation and distribution system, including: The first unit is used to obtain the scene data collected by the user terminal, construct a three-dimensional semantic map of the target scene based on the scene data; calculate the semantic entropy value of each area in the scene according to the three-dimensional semantic map, construct a scene semantic evolution tree based on the semantic entropy value, and obtain the semantic change trend of the target scene by updating the node state and transition weight in the scene semantic evolution tree in real time; The second unit is used to calculate the matching degree between the candidate content in the augmented reality content library and the semantic change trend, divide the relevant candidate content into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; The third unit is used to collect the interaction data between the user and the content group, where the interaction data includes interaction type, interaction duration, interaction result, and scene state change, establish an interaction behavior model based on the interaction data, and calculate the interaction response degree of each content group through the interaction behavior model; The fourth unit is used to classify the content groups according to the interaction response degree, determine the push priority based on the classification result, construct a directed push path graph in combination with the semantic association degree between the content groups, where the directed push path graph is used to guide the activation order of the content groups, and realize the smooth transition of the augmented reality experience through the ordered activation of the content groups.

[0014] In the third aspect of the embodiments of the present invention, a kind of electronic device is provided, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In the fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] The beneficial effects of this application are as follows: 1. By constructing a three-dimensional semantic map and a semantic evolution tree, the semantic change trend of the target scene can be reflected in real time, thereby improving the relevance and adaptability of augmented reality content.

[0017] 2. Through the interaction behavior model established based on user interaction data, the interaction response degree of the content group can be accurately evaluated, thereby optimizing the content push strategy and improving the user experience and satisfaction.

[0018] 3. Construct a directed push path graph by combining the semantic correlation degree of content groups, which can realize the orderly activation of content, ensure the smooth transition of the augmented reality experience, and enhance the user's immersion and participation. Brief Description of the Drawings

[0019] Figure 1 It is a schematic flowchart of the AI-driven interactive AR content intelligent generation and distribution method according to an embodiment of the present invention; Figure 2 It is a schematic diagram comparing the calculation accuracy of semantic entropy values in different scenarios according to an embodiment of the present invention; Figure 3 It is a complete flowchart of the processing mechanism based on semantic attributes and state transitions according to an embodiment of the present invention; Figure 4 It is a flowchart of the interaction behavior model and interaction response degree calculation according to an embodiment of the present invention; Figure 5 It is a schematic diagram comparing the preloading cache hit rates in different application scenarios according to an embodiment of the present invention. Detailed Embodiments

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0022] Figure 1 It is a schematic flowchart of the AI-driven interactive AR content intelligent generation and distribution method according to an embodiment of the present invention, as Figure 1 shown, the method includes: Obtain the scene data collected by the user terminal, and construct a three-dimensional semantic map of the target scene based on the scene data; calculate the semantic entropy value of each area in the scene according to the three-dimensional semantic map, construct a scene semantic evolution tree based on the semantic entropy value, and obtain the semantic change trend of the target scene by real-time updating the node state and transition weight in the scene semantic evolution tree; Calculate the matching degree between the candidate content in the augmented reality content library and the semantic change trend, divide the candidate content with relevance into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; Collect interaction data of the user with the content group, where the interaction data includes interaction type, interaction duration, interaction result, and scene state change. Based on the interaction data, establish an interaction behavior model, and calculate the interaction response degree of each content group through the interaction behavior model; Classify the content groups according to the interaction response degree, determine the push priority based on the classification result, and construct a directed push path graph in combination with the semantic association degree between the content groups. The directed push path graph is used to guide the activation order of the content groups, and realize the smooth transition of the augmented reality experience through the ordered activation of the content groups.

[0023] In an alternative implementation, calculate the semantic entropy value of each area in the scene according to the three-dimensional semantic map, construct a scene semantic evolution tree based on the semantic entropy value, and obtain the semantic change trend of the target scene by updating the node state and transition weight in the scene semantic evolution tree in real time, including: Divide the scene space of the three-dimensional semantic map into voxel grids, calculate the probability distribution of the appearance of each semantic category data in each voxel grid, and calculate the semantic entropy value of each voxel grid based on the probability distribution. The semantic entropy value is used to characterize the semantic complexity of the voxel grid; Obtain the neighborhood voxel set of each voxel grid, calculate the spatial distance between each voxel in the neighborhood voxel set and the target voxel, determine the spatial association weight based on the spatial distance, and perform weighted calculation on the semantic entropy value according to the spatial association weight to obtain the weighted semantic entropy value considering neighborhood influence; Construct a scene semantic evolution tree based on the weighted semantic entropy value, calculate the semantic state change amount, state transition probability, and state transition time interval between adjacent nodes in the scene semantic evolution tree, and use the semantic state change amount, the state transition probability, and the state transition time interval as the attributes of the connection edges between the nodes in the scene semantic evolution tree; Calculate the similarity between the nodes in the scene semantic evolution tree. The similarity is obtained through the weighted combination of the cosine similarity of the semantic state vectors, the Euclidean distance of the entropy value distribution matrix, and the time decay function. Dynamically update the scene semantic evolution tree based on the similarity to obtain the semantic change trend of the scene.

[0024] When dividing the scene space of the three-dimensional semantic map into voxel grids, adopt an adaptive voxel division strategy. Set the initial voxel size to 1 meter × 1 meter × 1 meter, and dynamically adjust the voxel size according to the scene complexity. For areas with rich details, the voxel size can be refined to 0.25 meter × 0.25 meter × 0.25 meter; for open and uniform areas, the voxel size can be enlarged to 2 meters × 2 meters × 2 meters.

[0025] For example, in an indoor scene, a voxel size of 0.5 m × 0.5 m × 0.5 m is used in the furniture-dense area, and a voxel size of 1 m × 1 m × 1 m is used in the corridor and hall areas. After the scene is divided, a total of 5,267 voxel units are generated.

[0026] When calculating the probability distribution of data of each semantic category in each voxel grid, the number of point clouds or the volume ratio of different semantic labels in each voxel is counted. The preset semantic categories include 20 common indoor object categories such as walls, floors, ceilings, doors, windows, tables, chairs, electrical appliances, etc. For voxel V1, it contains 45% of the floor point cloud, 30% of the table point cloud, 15% of the chair point cloud, and 10% of the point clouds of other objects, forming a probability distribution vector [0, 0.45, 0, 0, 0, 0.3, 0.15, ..., 0.1].

[0027] Based on the probability distribution, the semantic entropy value of each voxel grid is calculated. Using the Shannon entropy calculation formula, the sum of the product of the probability of each category and its logarithm is taken and then negated. For voxel V1, the calculated semantic entropy value is 1.58, indicating that the semantic complexity of this voxel is medium.

[0028] When obtaining the set of neighboring voxels of each voxel grid, a 26-neighborhood model is adopted, that is, all voxels (excluding the target voxel itself) within the 3×3×3 cube region around the target voxel are considered. For voxel V1, its set of neighboring voxels contains 26 voxels, which are respectively labeled as V2, V3, ..., V27. The spatial distance between each voxel in the set of neighboring voxels and the target voxel is calculated using the Euclidean distance formula, based on the center coordinates of the voxels. For example, the center coordinates of voxel V1 are (10.5, 8.2, 3.0), and the center coordinates of voxel V2 are (11.5, 8.2, 3.0), and the spatial distance between the two is 1.0.

[0029] When determining the spatial association weight based on the spatial distance, a Gaussian decay function is used, and the standard deviation parameter is set to 1.2. The closer the distance, the higher the weight. For voxels V1 and V2, the spatial distance is 1.0, and the calculated spatial association weight is 0.66. The semantic entropy value is weighted and calculated according to the spatial association weight, and the original entropy value of the target voxel and the weighted entropy values of the neighboring voxels are combined. For voxel V1, the original entropy value is 1.58, and the weighted semantic entropy value after considering the influence of the neighborhood is 1.72, indicating that the neighboring voxels increase the semantic complexity of the target area.

[0030] When constructing a scenario semantic evolution tree based on weighted semantic entropy values, the scenario state is used as a node and the state transition as an edge. Each node represents the semantic state of the scenario at a specific moment, containing two parts of information: the semantic state vector and the entropy value distribution matrix. The dimension of the semantic state vector is 128, which contains the semantic category distribution and spatial structure features of each voxel in the scenario; the size of the entropy value distribution matrix is consistent with the voxel grid divided by the scenario, and each element corresponds to the weighted semantic entropy value of a voxel. For example, node N1 represents the scenario state at time t1, its semantic state vector is [0.32, 0.45, 0.28, ..., 0.51], and the value at the position of voxel V1 in the entropy value distribution matrix is 1.72.

[0031] When calculating the semantic state change amount between adjacent nodes in the scenario semantic evolution tree, the Euclidean distance of the state vectors is used. For adjacent nodes N1 and N2, the semantic state vectors are [0.32, 0.45, 0.28, ..., 0.51] and [0.35, 0.42, 0.30, ..., 0.48] respectively, and the calculated semantic state change amount is 0.26, indicating that the difference degree between the two scenario states is moderate.

[0032] When calculating the state transition probability, it is based on historical state transition data and current state features. For the transition from node N1 to N2, 15 similar transitions were observed in the historical data, and the transition probability under the current conditions is 0.65. When calculating the state transition time interval, it combines the state difference degree and historical transition duration data. The average historical transition duration from node N1 to N2 is 2.3 seconds, and the predicted transition time considering the current state difference is 2.1 seconds.

[0033] The semantic state change amount, state transition probability, and state transition time interval are used as the attributes of the connection edges between nodes in the scenario semantic evolution tree. For example, edge E12 connects nodes N1 and N2, and its attribute values are (0.26, 0.65, 2.1), representing the semantic state change amount, state transition probability, and state transition time interval respectively. These edge attributes are used to guide scenario change prediction and resource allocation optimization.

[0034] When calculating the similarity between nodes in the scenario semantic evolution tree, the cosine similarity of the semantic state vector, the Euclidean distance of the entropy value distribution matrix, and the time decay function are comprehensively considered. The cosine similarity of the semantic state vector calculates the similarity degree of the semantic content between nodes; the Euclidean distance of the entropy value distribution matrix measures the entropy distribution difference between nodes; the time decay function considers the influence of time factors and decreases the similarity contribution as the time interval increases. For nodes N1 and N3, the cosine similarity of the semantic state vector is 0.78, the Euclidean distance of the entropy value distribution matrix is 0.45, the time interval is 5 seconds, and the time decay coefficient is 0.85. The comprehensively calculated similarity is 0.62.

[0035] The weighted combination of similarity calculation uses three weight coefficients, namely the semantic state vector weight of 0.5, the entropy value distribution matrix weight of 0.3, and the time decay weight of 0.2. The time decay function adopts an exponential decay model, and the decay rate is set to 0.05, so that the contribution of the state similarity 10 seconds ago is reduced by about 40%, and the contribution of the state similarity 20 seconds ago is reduced by about 65%.

[0036] When dynamically updating the scene semantic evolution tree based on similarity, node merging, edge weight adjustment, and path optimization operations are performed. Node merging rule: Nodes with a similarity exceeding 0.85 are merged into one node, and the attributes of the merged node are the weighted average of the original nodes; Edge weight adjustment rule: Update the state transition probability and time interval of the edge according to the newly added observation data; Path optimization rule: Retain the edges with a transition probability greater than 0.2, and delete the low-probability edges to simplify the tree structure. For example, it is found that the similarity between nodes N4 and N5 is 0.92, exceeding the threshold, and they are merged into a new node N45, and the state vector and entropy value distribution matrix are respectively taken as the weighted average.

[0037] Through the dynamically updated scene semantic evolution tree, the semantic change trend of the scene can be obtained. The change trends include: the main state transition path, the high-frequency transition mode, and the key state nodes. The main state transition path is a path sequence formed by connecting edges with high transition probabilities, such as N1→N2→N45→N7, and the transition probabilities are 0.65, 0.72, and 0.58 respectively; The high-frequency transition mode is a state cycle that repeats within a short period of time, such as the cycle from N3 to N6, and the mutual transition probabilities are 0.53 and 0.61; The key state node is a node that connects multiple high-probability transition edges, such as N2 connecting 3 edges with a transition probability greater than 0.6.

[0038] According to the predicted state transition path N1→N2→N45→N7, preload the visual resources corresponding to the nodes on the path in advance; For the high-frequency transition mode from N3 to N6, keep the resources in both states in the cache at the same time; For the key state node N2, preferentially allocate rendering resources and increase its preloading priority. The experimental results show that compared with the traditional method, the resource preloading strategy based on the semantic change trend reduces the resource loading waiting time by 42.7%, improves the scene switching smoothness by 38.5%, and the user experience score is improved by 25.3%.

[0039] Figure 2 Schematic diagram of the comparison of the semantic entropy value calculation accuracy in different scenarios of the embodiment of the present invention: This figure shows the comparison of the roaming experience accuracy rates of three different technical solutions under different scenario types. It is presented in the form of a line chart in the figure, where the horizontal axis represents five different complexity scenario types (simple indoor scenario, complex indoor scenario, open outdoor scenario, mixed urban scenario, dynamic multi-person scenario), and the vertical axis is the semantic entropy calculation accuracy rate (%). This technical solution (triangle marker) shows the best performance in various scenarios, ranging from 93% in the simple indoor scenario to 78% in the dynamic multi-person scenario. Although it decreases as the scenario complexity increases, the decline is the smallest. The method based on the clustering analysis (square marker) comes second, with the accuracy rate dropping from 85% to 60%. The traditional statistical method (round marker) performs the worst, dropping sharply from 80% to 55%. The performance differences among the three solutions are more obvious when the scenario complexity increases, especially in the mixed urban scenario and the dynamic multi-person scenario. This technical solution has obvious advantages compared with the other two methods, maintaining relatively high accuracy rates of 83% and 78% respectively, while the other methods decline significantly. This indicates that this technical solution has better adaptability and stability in complex and dynamic scenarios.

[0040] The existing technologies mainly adopt simple state diagrams or Markov models in the analysis of scene semantic changes, lacking quantitative analysis of scene semantic complexity and consideration of spatial relevance. Traditional methods usually only focus on the overall state transition of the scene, ignoring the semantic complexity differences and semantic change rules in different regions within the scene. The improvement starting point of this application is to introduce the semantic entropy theory to quantify the scene complexity, combine spatial correlation analysis to construct a more accurate scene semantic representation, and capture the dynamic change rules of scene semantics through the evolution tree model. Compared with the existing technologies, the prediction accuracy rate of scene changes in this application has increased by 37.2%, and the prediction error of state transition time has decreased by 43.5%, greatly improving the accuracy and timeliness of resource preloading, and providing key technical support for immersive interactive experiences.

[0041] In an optional implementation manner, constructing a scene semantic evolution tree based on the weighted semantic entropy value, and calculating the semantic state change amount, state transition probability, and state transition time interval between adjacent nodes in the scene semantic evolution tree includes: Extracting a set of semantic attributes of the scene based on the weighted semantic entropy value, where the set of semantic attributes includes object category, spatial position, object attributes, and relationships between objects. By performing threshold segmentation on the weighted semantic entropy value and extracting semantic regions, obtaining a semantic feature vector for each semantic region, and normalizing the semantic feature vector to obtain the attribute intensity value corresponding to each semantic attribute; Construct the node representation of the scenario semantic evolution tree according to the set of semantic attributes and the attribute intensity values. Each node contains a state vector and an entropy value distribution matrix. The state vector is constructed by organizing the attribute intensity values into a vector form according to a preset attribute sorting rule, and the entropy value distribution matrix is constructed by organizing the weighted semantic entropy values into a matrix form according to the spatial position relationship. Identify adjacent node pairs in the semantic evolution tree. For each adjacent node pair, calculate the difference between the components of the state vectors of the nodes, divide the difference by the maximum value of the corresponding component and introduce a smoothing factor to obtain a normalized semantic state change amount. Calculate the two-norm of the semantic state change amount to obtain the state distance. Take the negative of the state distance and perform an exponential operation to obtain the initial probability value. For each node, normalize the initial probability values of all its adjacent nodes to obtain the state transition base probability. Calculate the cosine similarity of the corresponding semantic attributes in the state vector to obtain the similarity value of each attribute. Weighted sum the similarity value and a preset importance weight to obtain the state transition time interval.

[0042] When extracting the set of semantic attributes of the scenario based on the weighted semantic entropy value, perform multi-level threshold segmentation on the weighted semantic entropy value. Set the initial threshold to 0.35, the step size to 0.1, and a total of three levels of thresholds: 0.35, 0.45, and 0.55. Divide the weighted semantic entropy value according to these three levels of thresholds to obtain multiple semantic regions.

[0043] For a certain scenario image, 5 semantic regions are obtained through threshold segmentation, which are respectively labeled as R1, R2, R3, R4, and R5. Apply a feature extractor to each semantic region to obtain a semantic feature vector. Feature extraction uses a deep convolutional neural network, which includes 8 convolutional layers and 3 fully connected layers, and outputs a feature vector with a dimension of 128.

[0044] For example, the original feature vector extracted from region R1 is [0.75, 0.23, 0.56, ..., 0.41]. Normalize the extracted semantic feature vector so that each component value falls between 0 and 1 to obtain the attribute intensity value corresponding to each semantic attribute. The normalization uses the maximum-minimum normalization method. For region R1, the normalized attribute intensity value vector is [0.82, 0.15, 0.64, ..., 0.53].

[0045] The set of semantic attributes includes four categories: object category, spatial location, object attributes, and relationships between objects. The object category attribute contains 64 predefined common object categories, such as people, vehicles, buildings, etc. Each category corresponds to an attribute intensity value, representing the confidence that the area contains objects of that category. In region R1, the attribute intensity value of the "person" category is 0.92, and the attribute intensity value of the "building" category is 0.18.

[0046] The spatial location attribute is represented by relative coordinates, including the center coordinates (x, y) and size information (width, height), normalized to the range of 0 - 1. The spatial location attribute value of region R1 is [0.35, 0.42, 0.28, 0.31], indicating that the center is located at the position (0.35, 0.42) of the image, the width is 0.28 of the image width, and the height is 0.31 of the image height.

[0047] Object attributes include characteristics such as color, texture, and shape, represented by a multi-dimensional vector. In the object attribute value of region R1, the color feature is [0.75, 0.25, 0.45], representing the intensities of the red, green, and blue channels. The relationship attribute between objects describes the spatial and semantic associations between different regions, including relative position relationships and interaction relationships. The relationship attribute value between region R1 and R2 is [0.8, 0.1, 0.05, 0.05], representing the probability distribution of the four relationships: "contains", "adjacent", "overlaps", and "separated".

[0048] When constructing the node representation of the scene semantic evolution tree based on the set of semantic attributes and the attribute intensity values, each node contains two parts of information: a state vector and an entropy value distribution matrix. When constructing the state vector, the attribute intensity values are organized into a vector form according to a preset attribute sorting rule. The sorting rule is: first object category, then spatial location, then object attributes, and finally relationships between objects.

[0049] For a specific scene node N1, the first 64 components of its state vector represent the object category intensity values, such as [0.92 (person), 0.05 (vehicle), 0.18 (building),...]; the next 4 components represent the spatial location information [0.35, 0.42, 0.28, 0.31]; then the object attribute information [0.75, 0.25, 0.45,...]; and finally the relationship information between objects [0.8, 0.1, 0.05, 0.05,...].

[0050] When constructing the entropy value distribution matrix, the weighted semantic entropy values are organized into a matrix form according to the spatial position relationship. The matrix size is the same as that of the original scene image, and each element represents the semantic entropy value at the corresponding position. For node N1, the size of its entropy value distribution matrix is 16×16, and some values are [[0.42, 0.38, 0.51, ...], [0.37, 0.43, 0.48, ...], ...].

[0051] When identifying adjacent node pairs in the semantic evolution tree, a strategy combining temporal adjacency and spatial similarity is adopted. Temporal adjacency means that two nodes are adjacent in the time series, such as frame t and frame t+1; spatial similarity is judged by calculating the cosine similarity of the node state vectors, and the similarity threshold is set to 0.85. Node pairs with a similarity exceeding the threshold are considered to have spatial similarity.

[0052] For the identified adjacent node pair (N1, N2), calculate the difference between the components of the state vectors between the nodes. For example, the state vector of N1 is [0.92, 0.05, 0.18, ...], and the state vector of N2 is [0.88, 0.08, 0.22, ...]. The calculated difference vector is [0.04, -0.03, -0.04, ...]. Divide the difference by the maximum value of the corresponding component and introduce a smoothing factor to obtain the normalized semantic state change amount. The smoothing factor is set to 0.05 to avoid the case of division by zero. For the first component 0.04 of the difference vector, the corresponding maximum value is 0.92, and the normalized change amount is 0.04 / (0.92 + 0.05) = 0.041.

[0053] When calculating the state distance by taking the two-norm of the semantic state change amount, the method of taking the square root of the sum of squares is adopted. For the normalized semantic state change amount vector [0.041, -0.032, -0.042, ...], calculate the sum of the squares of each component and then take the square root to obtain a state distance value of 0.217. Take the negative of the state distance and perform an exponential operation to obtain the initial probability value.

[0054] For the state distance 0.217, after taking the negative, it is -0.217, and the exponential operation gives an initial probability value of 0.805. For each node, normalize the initial probability values of all its adjacent nodes to obtain the basic probability of state transition. For example, node N1 has three adjacent nodes N2, N3, and N4, and the corresponding initial probability values are 0.805, 0.723, and 0.652 respectively. The normalized state transition probabilities are 0.37, 0.33, and 0.30 respectively.

[0055] When calculating the cosine similarity of the corresponding semantic attributes in the state vector to obtain the similarity values of each attribute, the cosine similarities of four types of semantic attributes, namely object category, spatial position, object attributes, and relationships between objects, are calculated respectively. For the node pair (N1, N2), the object category similarity is 0.95, the spatial position similarity is 0.92, the object attribute similarity is 0.87, and the relationship similarity between objects is 0.81.

[0056] The similarity values are weighted and summed with the preset importance weights to obtain the state transition time interval. The importance weights of the four types of attributes are set to 0.4, 0.3, 0.2, and 0.1 respectively. The total similarity obtained by weighted summation is 0.95×0.4 + 0.92×0.3 + 0.87×0.2 + 0.81×0.1 = 0.911. The total similarity is converted into the state transition time interval, using an inverse proportional relationship. The base value of the time interval is 1.0, and the actual interval is the base value divided by the total similarity, that is, 1.0 / 0.911 = 1.098, indicating that the expected time from state N1 to state N2 is 1.098 time units.

[0057] Figure 3 The complete flowchart of the processing mechanism based on semantic attributes and state transitions in the embodiments of the present invention is as follows: This flowchart describes a processing mechanism based on semantic attributes and state transitions. First, a set of semantic attributes of the scene is extracted by weighted semantic entropy values, including elements such as object category, spatial position, object attributes, and relationships between objects. After threshold segmentation of the weighted semantic entropy, semantic regions are obtained, and semantic feature vectors are extracted for each semantic region. Through normalization processing, the attribute intensity values corresponding to each semantic attribute are obtained. Secondly, based on the set of semantic attributes and the attribute intensity values, node representations of the scene semantic evolution tree are constructed. Each node contains a state vector and an entropy value distribution matrix. The attribute intensity values are organized into a vector form according to preset rules to construct the state vector, and the entropy value distribution matrix is constructed in a matrix form according to the spatial position relationship. Then, adjacent node pairs in the semantic evolution tree are identified, and the differences between the components of the state vectors between the nodes are calculated. The differences are divided by the maximum value of the corresponding components and a smoothing factor is introduced to obtain the normalized semantic state change amount. Next, the two-norm of the semantic state change amount is calculated to obtain the state distance, and the exponential operation is performed on the state distance to obtain the initial probability value. For each node, the initial probability values of all its adjacent nodes are normalized to obtain the state transition base probability. Finally, the cosine similarity of the corresponding semantic attributes in the state vector is calculated to obtain the similarity values of each attribute. The similarity values are weighted and summed with the preset importance weights to obtain the state transition time interval. The entire process realizes the accurate expression and smooth conversion of semantic states through rigorous mathematical processing.

[0058] In the prior art, simple state transition diagrams or Markov chain models are mainly adopted in the analysis of scene semantic evolution. These methods usually only consider the overall state of the scene, ignoring the fine-grained changes and complex associations of the semantic elements within the scene. Traditional methods such as rule-based scene understanding systems or statistical state transition models are difficult to accurately capture the dynamic evolution laws of object categories, spatial positions, object attributes, and relationships between objects in the scene. The improvement starting point of this application is to construct a multi-level and fine-grained scene semantic evolution tree, integrating the semantic entropy theory and attribute intensity modeling to achieve precise quantification of the scene semantic state and its changes.

[0059] The main improvements of this application compared with the prior art include: introducing a weighted semantic entropy value as a measure of the information amount of scene content to achieve adaptive recognition of important regions in the scene; using a multi-dimensional semantic attribute set to represent scene content, including object categories, spatial positions, object attributes, and relationships between objects, providing a more comprehensive scene description; designing a dual-node representation of a state vector and an entropy value distribution matrix to capture both the structured semantic information and the unstructured entropy distribution characteristics of the scene; proposing a state transition metric method based on the normalized semantic state change amount to accurately quantify the difference degree between scene states; establishing a state transition time model based on multi-attribute similarity weighting to achieve accurate prediction of the conversion timing characteristics.

[0060] In an alternative embodiment, calculating the matching degree between candidate content in the augmented reality content library and the semantic change trend, dividing the relevant candidate content into content groups according to the matching degree, and generating corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map includes: Collecting the feature information of candidate content in the augmented reality content library, converting the feature information into feature vectors, where the feature vectors include visual feature components, semantic feature components, and interaction feature components; Performing a matching calculation between the feature vector of the candidate content and the semantic change trend, and obtaining the matching degree score of each candidate content by calculating the consistency degree of the change direction of the feature vector of the candidate content in the spatial dimension, attribute dimension, and interaction dimension with the semantic change trend; Taking the vector inner product of the matching degree scores as the similarity value between the candidate contents, constructing a similarity matrix; calculating the mean and standard deviation of the similarity matrix, setting a dynamic clustering threshold based on the mean and the standard deviation, and dividing the candidate contents with a similarity higher than the dynamic clustering threshold into the same content group; Extracting scene features from the three-dimensional semantic map, where the scene features include spatial topological structure, semantic attribute distribution, and interaction hot spot distribution, and generating corresponding trigger conditions for each content group based on the scene features; Monitor the scene changes in the 3D semantic map. When the scene changes meet the trigger conditions of the content group, calculate the update metrics for the corresponding content group, where the update metrics combine the access frequency, timeliness, and scene relevance of the content group; Determine the update priority of the content group according to the update metrics. When the update priority exceeds the preset priority threshold, select the content with the highest matching degree with the current semantic change trend from the candidate content library as the update strategy.

[0061] Extract the feature information of the candidate content from the augmented reality content library. The feature information includes visual features, semantic features, and interaction features. Visual features can be obtained through image processing techniques, such as using a convolutional neural network (CNN) to extract the color histogram, texture features, etc. of an image. Semantic features are analyzed from the description text of the content through natural language processing techniques to extract keywords and their semantic relationships. Interaction features are obtained through user interaction data analysis, such as user click-through rate, dwell time, etc.

[0062] Convert this feature information into feature vectors. The feature vector should contain three components: a visual feature component, a semantic feature component, and an interaction feature component. The value of each component can be normalized to make comparisons on the same scale.

[0063] Calculate the matching degree between the feature vector of the candidate content and the semantic change trend. By analyzing the change directions of the feature vector of the candidate content in the spatial dimension, attribute dimension, and interaction dimension, evaluate its consistency with the semantic change trend. Specifically, the matching degree can be judged by comparing the change rate and trend direction of the feature vector.

[0064] If the visual feature vector of a certain candidate content is consistent with the current popular trend in terms of color, then the matching degree score of this content will be relatively high. This score can be obtained by comprehensively considering the change situations in each dimension.

[0065] Organize the matching degree scores of all candidate contents into a vector, and calculate the similarity between these scores. The similarity can be obtained by calculating the inner product of the score vectors. After constructing the similarity matrix, calculate the mean and standard deviation of this matrix. Based on the mean and standard deviation, set a dynamic clustering threshold, and divide the candidate contents with similarity higher than this threshold into the same content group.

[0066] If the mean of the matching degree scores of a group of candidate contents is 0.8 and the standard deviation is 0.1, then a threshold of 0.9 can be set, and the contents with scores higher than 0.9 are grouped together.

[0067] Extract scene features from the 3D semantic map, mainly including spatial topological structure, semantic attribute distribution, and interaction hotspot distribution. The spatial topological structure can be obtained by analyzing the relative positions and connection relationships of objects in the scene. The semantic attribute distribution is statistically analyzed through the semantic labels of objects in the scene. The interaction hotspot distribution is analyzed through the interaction behavior data of users in the scene to identify the main areas of user attention.

[0068] Based on the extracted scene features, generate corresponding trigger conditions for each content group. For example, if the scene features of a certain content group show that the user's interaction frequency in a specific area is relatively high, the change in this area can be set as the trigger condition.

[0069] Monitor the scene changes in the 3D semantic map. When the scene changes meet the trigger conditions of a certain content group, calculate the update metrics for this content group. The update metrics should be comprehensively evaluated by combining the access frequency, timeliness, and scene relevance of the content group. For example, if the access frequency of a certain content group has increased significantly in the past week and is highly relevant to the current scene changes, its update priority will be correspondingly increased.

[0070] According to the calculated update metrics, determine the update priority of each content group. When the update priority of a certain content group exceeds the preset priority threshold, select the content with the highest matching degree with the current semantic change trend from the candidate content library as the update strategy. The selection of this content should ensure that it can effectively meet the user's needs and improve the user experience.

[0071] For example, if the update priority of a certain content group is 95 and the score of the candidate content with the highest matching degree is 0.9, this candidate content can be used as the update strategy for pushing.

[0072] In an alternative implementation manner, an interaction behavior model is established based on the interaction data, and calculating the interaction response degree of each content group through the interaction behavior model includes: Extract the interaction time-series features, interaction spatial features, and interaction modality features from the interaction data, and fuse the interaction time-series features, the interaction spatial features, and the interaction modality features to construct a behavior feature vector; Establish an interaction behavior model based on the behavior feature vector. The interaction behavior model adopts a hybrid structure of a recurrent neural network and an attention mechanism. Capture the time-series dependencies of the interaction behavior model through the recurrent neural network, and use the attention mechanism to highlight the importance of interaction features to generate a dynamic representation of the interaction behavior model; Calculate the interaction response degree of the content group based on the interaction behavior model, perform similarity matching between the feature vector of the content group and the dynamic representation, and dynamically weight the result of the similarity matching through a temporal smoothing factor to obtain a response score representing the matching degree between the content group and the interaction behavior.

[0073] When extracting the interaction temporal features, interaction spatial features, and interaction modality features from the interaction data, parse and extract features from the interaction logs of the user and the content group. Interaction temporal features include time dimension information such as the time point when the interaction occurs, the interaction duration, and the interaction interval time. For example, record that the user clicks on content group A at 10:23:45 on March 15, 2025, and stays for 45 seconds; clicks on content group B at 10:25:12 and stays for 30 seconds, obtaining a time stamp sequence, a duration sequence, and an interaction interval sequence. Interaction spatial features include spatial dimension information such as the physical location where the interaction occurs, the device posture, and the contact coordinates. For example, the user clicks on content group C at the coordinate position (320, 450), the device is in the landscape screen state, and the contact pressure value is 0.7. Interaction modality features include the type of interaction method, such as features of different interaction modes such as click, slide, voice command, and eye movement tracking. For example, the user uses a long press operation on content group D, the pressure value is 0.85, and the duration is 2 seconds.

[0074] Fuse the above-extracted interaction temporal features, interaction spatial features, and interaction modality features to construct a behavior feature vector. Feature fusion uses a multi-layer perceptron structure. First, perform normalization processing on different types of features, and then map them to a common feature space through a fully connected layer. In specific implementation, the temporal features are mapped to a 128-dimensional vector, the spatial features are mapped to a 64-dimensional vector, the modality features are mapped to a 32-dimensional vector, and then concatenated to form a 224-dimensional joint representation. Finally, a final behavior feature vector is generated through a fully connected layer with a dimension of 256. For an interaction between the user and content group E, the extracted temporal feature values may be [0.25, 0.78, 0.45,...], the spatial feature values are [0.62, 0.31, 0.58,...], the modality feature values are [0.9, 0.2, 0.4,...], and the fused behavior feature vector is [0.53, 0.71, 0.38, 0.42,..., 0.65].

[0075] When establishing an interaction behavior model based on behavior feature vectors, a hybrid structure of a recurrent neural network and an attention mechanism is adopted. The recurrent neural network part uses long short-term memory (LSTM) units, which include an input layer, two hidden layers, and an output layer. The input layer receives a sequence of behavior feature vectors. Each LSTM unit contains 64 neurons and is used to capture the temporal dependencies of interaction behaviors. For example, the user first quickly browses content group F and then returns to carefully read content group G, and this kind of behavior sequence pattern can be learned by the LSTM unit. The attention mechanism adopts a self-attention structure to calculate the correlation score between behavior feature vectors and assigns higher weights to important features. For example, in the scenario of reading an article, the weight of the dwell time feature may be 0.6, while in the scenario of browsing pictures, the weight of the click position feature may be 0.7. Through the attention mechanism, the importance of each feature in different scenarios can be adaptively adjusted.

[0076] During the training process of the interaction behavior model, historical interaction data is used to construct a training set, with the behavior feature vector sequence as the input and the user's subsequent interaction behavior as the prediction target. The training uses the backpropagation algorithm, the loss function is the cross-entropy loss, the optimizer is the Adam optimizer, the initial value of the learning rate is set to 0.001, and a learning rate decay strategy is adopted, with a 10% reduction every 10 training rounds. The training iterates 500 rounds, and the prediction accuracy on the validation set reaches 85.6%. After training, the model can generate a dynamic representation vector representing the user's current interaction intention based on the sequence of the user's historical interaction behavior feature vectors.

[0077] When calculating the interaction response degree of a content group through the interaction behavior model, the feature vector of the content group to be evaluated is matched with the dynamic representation generated by the model for similarity. The content group feature vector includes attributes such as content theme, format type, and complexity, and adopts the same dimensional representation method as the behavior feature vector. The similarity matching uses the cosine similarity calculation method, and the cosine angle between the content group feature vector and the dynamic representation vector is used as the original matching score. For example, the feature vector of content group H is [0.65, 0.28, 0.73,...], and the current dynamic representation vector is [0.70, 0.31, 0.68,...], and the calculated cosine similarity is 0.91, indicating that content group H highly matches the current interaction behavior.

[0078] To eliminate the influence of short-term interaction fluctuations, a temporal smoothing factor is introduced to dynamically weight the similarity matching results. The temporal smoothing factor is based on the exponential moving average algorithm, which assigns decreasing weights to the matching scores of the most recent N interactions to obtain a smoothed responsiveness score. In the calculation of the smoothing factor, the decay rate α is set to 0.8, and the weights of the most recent 5 interactions are 0.8^0 = 1, 0.8^1 = 0.8, 0.8^2 = 0.64, 0.8^3 = 0.512, 0.8^4 = 0.4096 respectively. For content group I, the original matching scores of the most recent 5 interactions are 0.85, 0.78, 0.82, 0.79, 0.73 respectively. After weighted averaging, the responsiveness score is 0.81.

[0079] In the application strategy of the responsiveness score, a high response threshold is set to 0.75 and a medium response threshold is set to 0.5. Content groups with a responsiveness score higher than 0.75 are marked as "high response" and have the highest priority; content groups with scores between 0.5 and 0.75 are marked as "medium response"; content groups with scores lower than 0.5 are marked as "low response". For content group I with a responsiveness score of 0.81, it is marked as "high response" and obtains priority processing rights in resource allocation and content presentation.

[0080] To verify the effectiveness of the interactive responsiveness calculation, 100 users are tested. Each user interacts with the system for 30 minutes, generating a total of 28,634 interaction records. The analysis results show that users tend to have more subsequent interactions with "high response" content groups. The click-through rate of "high response" content groups is 2.7 times higher than that of "low response" content groups, and the stay time is 3.2 times longer, proving that the interactive responsiveness calculation method can effectively predict users' interactive interests.

[0081] Figure 4 The following is the flowchart of the interactive behavior model and interactive responsiveness calculation for the embodiments of the present invention: This flowchart shows a complete processing flow for feature extraction and responsiveness calculation based on interaction data. First, three types of features are extracted from the interaction data: aggregated interaction features, interaction space features, and interaction scenario features. These feature relationships are constructed into behavior feature warnings, and then divided into two branches for processing: further, the mutual dependencies and associated behavior patterns between neural network interactions are derived; then, the importance of prominent interaction features is analyzed through the focus mechanism, and the processing results of these two key branches act together to generate a dynamic representation process of the interactive behavior model. Finally, the interactive responsiveness is calculated through this representation. This processing method combining multi-dimensional feature extraction and deep learning not only ensures accurate modeling of the interactivity of interactive behaviors but also achieves precise capture of key interactive behavior features, thereby enabling more accurate evaluation of the interaction effect between users and content. The whole constitutes a complete closed-loop system from the original interaction data to the final responsiveness calculation.

[0082] The prior art mainly relies on static feature matching or simple collaborative filtering methods when calculating content responsiveness, and cannot adapt to complex and changeable interaction scenarios. Traditional methods such as rule-based scoring systems or statistic-based recommendation algorithms often ignore the temporal and multi-modal characteristics of interaction behaviors, resulting in inaccurate responsiveness calculation. The improvement starting point of this application is to construct a behavior model that can capture the dynamic characteristics of interaction behaviors, and integrate temporal features, spatial features, and modal features for comprehensive analysis. By introducing a recurrent neural network to capture long-term dependencies and combining an attention mechanism to highlight key features, accurate modeling of complex interaction behaviors is achieved. Compared with the prior art, the prediction accuracy of interaction behaviors in this application has increased by 23.5%, and the adaptability to different interaction modes has increased by 18.7%, significantly improving the accuracy and stability of content responsiveness evaluation, providing a reliable basis for the precise preloading and real-time rendering of content groups, and optimizing the user interaction experience.

[0083] In an optional implementation manner, a directed push path graph is constructed in combination with the semantic association degree between the content groups, and the directed push path graph is used to guide the activation order of the content groups. The smooth transition of the augmented reality experience achieved by the ordered activation of the content groups includes: Normalize the concept weights, entity vectors, and context features in the content group to obtain a normalized feature vector, and construct a semantic representation of the content group based on the normalized feature vector; Calculate the centrality, density, and betweenness centrality of the content group, and obtain the node importance according to the weighted combination of the centrality, the density, and the betweenness centrality; construct a directed push path graph based on the node importance, and the directed push path graph is used to represent the activation dependency relationship between the content groups; In the directed push path graph, calculate the activation path score by using a conditional probability function considering historical states. The conditional probability function takes the semantic association strength, edge weight, and resource consumption of adjacent content groups as input parameters, and selects the path with the highest score as the optimal activation path; According to the access order of the optimal activation path, preload the content groups to be activated into the corresponding cache levels, calculate the content difference degree between adjacent content groups based on the semantic representation of the preloaded content groups, and calculate the context relevance in combination with real-time scene information; Input the content difference degree and the context relevance into an adaptive state smoothing function to dynamically generate the transition time between adjacent content groups, and control the activation process of the content groups according to the transition time to achieve smooth transition of states.

[0084] Normalize the concept weights, entity vectors, and context features in the content group. The purpose of this step is to ensure that different features can be compared on the same scale. Specifically, the concept weights can be normalized by subtracting the minimum value of each concept's weight from the maximum value and dividing by the difference between the maximum and minimum values. The entity vectors and context features are also processed in the same way, and the final standardized feature vectors will be used for subsequent semantic representation.

[0085] After obtaining the standardized feature vectors, it is necessary to construct the semantic representation of the content group based on these vectors. The weighted average method can be used to weight different feature vectors according to their importance to generate a comprehensive semantic representation. This representation will reflect the overall characteristics of the content group and facilitate subsequent analysis and processing.

[0086] Calculate the centrality, density, and betweenness centrality of the content group to evaluate the importance of each node. The centrality can be determined by analyzing the number of connections of the node in the network, the density represents the tightness of the connections between nodes, and the betweenness centrality reflects the key role of the node in information flow. By combining these three indicators with weights, the importance score of the node is obtained, providing a basis for constructing the directed push path graph.

[0087] Based on the node importance, construct a directed push path graph. This graph will represent the activation dependence relationship between content groups. Each node represents a content group, and the direction of the edge indicates the activation order. The construction of this graph needs to ensure that nodes with high importance are activated first to achieve effective content push.

[0088] In the directed push path graph, a conditional probability function considering the historical state is used to calculate the score of the activation path. The inputs of this function include the semantic association strength, edge weight, and resource consumption of adjacent content groups. By analyzing these factors, the path with the highest score is selected as the optimal activation path. This process ensures the rationality and efficiency of the activation order.

[0089] According to the access order of the optimal activation path, preload the content groups to be activated into the corresponding cache levels. The preloaded content groups should calculate the content difference degree between adjacent content groups based on their semantic representations. The calculation of the content difference degree needs to consider the theme, style, and other relevant features of the content to ensure the coherence between adjacent content groups.

[0090] Combine real-time scenario information to calculate the context relevance. The context relevance reflects the impact of the current environment on the activation of content groups and can help adjust the activation strategy. By analyzing the user's behavior patterns and current environmental characteristics, dynamically adjust the content push strategy.

[0091] Input the content difference degree and context relevance into the adaptive state smoothing function to dynamically generate the transition time between adjacent content groups. This function will adjust the activation process of the content groups according to the differences between the contents and the changes in the environment to achieve a smooth transition of the state. The setting of the transition time needs to consider the user experience and avoid overly abrupt content switching.

[0092] Control the activation process of the content groups according to the dynamically generated transition time. By precisely regulating the activation timing of each content group, ensure the fluency and coherence of the user in the augmented reality experience. This process needs to monitor the user feedback in real time so as to adjust the activation strategy in a timely manner and improve the user satisfaction.

[0093] Suppose there are three content groups A, B, and C, representing different augmented reality scenarios respectively. The weights of content group A is 0.8, B is 0.6, and C is 0.4. After normalization, the normalized feature vectors of A, B, and C are [0.8, 0.7, 0.6], [0.6, 0.5, 0.4], [0.4, 0.3, 0.2] respectively. According to the calculated centrality, density, and betweenness centrality, assume the node importance is A: 0.9, B: 0.7, C: 0.5. The finally constructed directed push path graph will preferentially activate content group A, followed by B and C.

[0094] In an alternative implementation, according to the access order of the optimal activation path, preload the content groups to be activated into the corresponding cache levels, calculate the content difference degree between adjacent content groups based on the semantic representation of the preloaded content groups, and calculate the context relevance in combination with the real-time scene information, including: Construct the temporal feature vector of the content group, where the temporal feature vector includes access frequency features, time interval features, and duration features, and generate the dynamic access pattern of the content group based on the temporal feature vector; Calculate the preloading priority score according to the dynamic access pattern, where the preloading priority score is obtained by the weighted combination of the access frequency score, time urgency, and resource requirement degree, and determine the loading order of the content groups based on the preloading priority score; Perform cache level allocation on the content groups whose loading order is determined. Determine the optimal cache level by calculating the difference between the utility value and the storage cost of the content group in each cache level, and load the content group into the corresponding cache level to generate a cache allocation scheme; Obtain the multimodal features of the content groups that have completed cache allocation, perform weighted fusion on the multimodal features to obtain a fusion feature matrix, and generate a semantic embedding vector based on the fusion feature matrix through a non-linear mapping function; Collect the scenario state information, user behavior information, and environmental information corresponding to the semantic embedding vector, input the scenario state information, the user behavior information, and the environmental information into a context encoding function to generate a context feature representation, and calculate the context relevance between content groups according to the context feature representation.

[0095] When constructing the time series feature vector of a content group, collect the access record data of each content group, including the access timestamp list and the access duration list. For content group C_i, count the number of accesses within the past T time window to obtain the access frequency feature f_i; calculate the time difference sequence between two adjacent accesses to obtain the time interval feature i_i; record the access duration sequence of each access to obtain the access duration feature d_i. Combine these three types of features to form a time series feature vector [f_i, i_i, d_i].

[0096] When generating the dynamic access pattern of a content group based on the time series feature vector, adopt a sliding time window method to divide the continuous time series into multiple overlapping time windows. Extract pattern features from the access behaviors within each time window, such as the periodic change of access frequency, peak access time period, typical access sequence, etc. By comparing the access patterns of different time windows, identify the stable patterns and change trends to form a description of the dynamic access pattern of the content group. For example, the access frequency of content group C_1 is significantly higher during the period from 9:00 to 11:00 on weekdays than other periods, and it is usually accessed before accessing C_2.

[0097] When calculating the preloading priority score according to the dynamic access pattern, first calculate the access frequency score S_f, which reflects the possibility of the content group being accessed; secondly calculate the time urgency S_t, which represents the time urgency of the expected next access; finally calculate the resource requirement S_r, which represents the amount of resources required to load this content group. The preloading priority score P is obtained through weighted combination: P = w_f×S_f + w_t×S_t + w_r×S_r, where w_f, w_t, and w_r are the weight coefficients of the three factors, and satisfy w_f + w_t + w_r = 1.

[0098] Taking content group C_1 as an example, if its access frequency score S_f = 0.85, time urgency S_t = 0.92, resource requirement S_r = 0.4, weight coefficients w_f = 0.4, w_t = 0.4, w_r = 0.2, then the preloading priority score P = 0.4×0.85 + 0.4×0.92 + 0.2×0.4 = 0.78. According to the calculated preloading priority scores, sort all content groups from high to low according to the scores to determine the loading order of the content groups.

[0099] When performing cache - level allocation for content groups with a determined loading order, assume that the system has n cache levels {L_1, L_2, ..., L_n}, and each cache level has different access speeds and storage capacities.

[0100] For each content group C_i and each cache level L_j, calculate the utility value U_ij, which represents the performance gain brought by storing C_i in L_j; at the same time, calculate the storage cost C_ij, which represents the cost of occupying the space of L_j. By calculating the difference D_ij = U_ij - C_ij between the utility value and the storage cost, determine the optimal cache level for each content group.

[0101] The utility values of content group C_2 in the three - level cache are U_21 = 0.95, U_22 = 0.75, U_23 = 0.45 respectively, and the storage costs are C_21 = 0.8, C_22 = 0.5, C_23 = 0.2 respectively. Then the differences are D_21 = 0.15, D_22 = 0.25, D_23 = 0.25. Therefore, select L_2 or L_3 as the cache level for C_2. Considering the cache - level allocation results of all content groups comprehensively, generate an overall cache allocation scheme.

[0102] When obtaining the multi - modal features of content groups that have completed cache allocation, extract the text features, image features, audio features, etc. of the content groups. For text features, use a text encoder to extract semantic representations; for image features, use a visual feature extractor to obtain visual representations; for audio features, use an audio feature extractor to obtain acoustic representations. Perform weighted fusion on the extracted multi - modal features to obtain a fused feature matrix M.

[0103] For content group C_3, its text feature vector is [0.5, 0.8, 0.3], its image feature vector is [0.7, 0.2, 0.9], and its audio feature vector is [0.4, 0.6, 0.5]. Use the weight coefficients [0.5, 0.3, 0.2] for weighted fusion to obtain a fused feature vector [0.54, 0.56, 0.52]. Based on the fused feature matrix M, generate a semantic embedding vector S through a non - linear mapping function. The non - linear mapping function adopts a multi - layer perceptron structure to map high - dimensional features to a low - dimensional semantic space and capture the essential semantic information of the content group.

[0104] When collecting scenario state information, user behavior information, and environmental information corresponding to the semantic embedding vector, the scenario state information includes the current application state, active page type, etc.; the user behavior information includes the user operation sequence, dwell time, click behavior, etc.; the environmental information includes the network state, device performance, battery level, etc. These information are input into the context encoding function to generate the context feature representation C. The context encoding function adopts the attention mechanism to assign different weights to different types of context information, highlighting the key factors related to the current scenario.

[0105] According to the context feature representation C, calculate the context relevance R between content groups. For the content group pair (C_i, C_j), obtain the context relevance R_ij by calculating the conditional similarity of their semantic embedding vectors S_i and S_j under the context feature representation C. For example, in the scenario where the user browses the electronic product page, the context relevance between the product description content group and the product review content group is 0.86, while the context relevance between the product description content group and the payment process content group is only 0.35.

[0106] Based on the calculated content difference degree and context relevance, the system can predict the most likely content group sequence that the user will access, and optimize the preloading strategy accordingly to improve the resource utilization efficiency and user experience. For example, when it is detected that the user is browsing the details page of product A and the dwell time exceeds 30 seconds, the system predicts that the user may view the product review next, so the review content group is preloaded into the L_1 cache in advance, enabling the user to immediately see the review content after clicking the "View Review" button without waiting for the loading process.

[0107] Figure 5 Schematic diagram of the comparison of preloading cache hit rates in different application scenarios of the embodiments of the present invention: The figure shows the comparison of cache hit ratios of four different technical solutions under seven application scenarios. The horizontal axis in the figure represents different application scenarios, including VR exhibition halls, large open-world games, interactive educational applications, smart home control, multi-person collaboration environments, high-precision medical visualization, and autonomous driving simulations; the vertical axis is the cache hit ratio (%). This technical solution (triangle mark) performs optimally in each scenario, with the hit ratio always remaining between 87 - 96%, reaching the highest value of 96% in the smart home control scenario. The semantic-driven cache (plus mark) comes second, with the hit ratio fluctuating between 75 - 88%. The performance of the preset path preloading (square mark) is in the middle, with the hit ratio between 68 - 83%. The LRU cache algorithm (circle mark) performs the worst, with the hit ratio dropping to about 59% at the lowest. The four solutions show similar fluctuation trends in different application scenarios, but this technical solution has the smallest fluctuation range, indicating better stability and adaptability. Especially in computationally intensive scenarios such as high-precision medical visualization and autonomous driving simulations, this technical solution can still maintain a high hit ratio, showing obvious performance advantages.

[0108] Existing technologies usually adopt simple strategies such as LRU (Least Recently Used) or LFU (Least Frequently Used) for content preloading and cache management, only considering access frequency and time factors and ignoring content semantics and context information. The improvement starting point of this application is to integrate the temporal characteristics, semantic characteristics, and context information of content groups to achieve more accurate preloading decisions. The specific improvements include: constructing a multi-dimensional temporal feature vector containing access frequency, time interval, and duration to capture richer access patterns; introducing multi-modal feature fusion and semantic representation to understand the essential semantics of content groups; combining real-time scene states, user behaviors, and environmental information to calculate context relevance and implement a context-aware preloading strategy. Through these improvements, this application significantly improves the precision and recall rate of content preloading. Experimental results show that compared with traditional methods, the preloading hit ratio has increased by 35%, the resource utilization efficiency has increased by 28%, and the user waiting time has decreased by 43%, providing a smoother interaction experience for users.

[0109] In the second aspect of the embodiments of the present invention, An AI-driven interactive AR content intelligent generation and distribution system is provided, including: A first unit for obtaining scene data collected by a user terminal, constructing a three-dimensional semantic map of a target scene based on the scene data; calculating the semantic entropy value of each region in the scene according to the three-dimensional semantic map, constructing a scene semantic evolution tree based on the semantic entropy value, and obtaining the semantic change trend of the target scene by updating the node states and transition weights in the scene semantic evolution tree in real time; A second unit, configured to calculate the matching degree between candidate content in an augmented reality content library and the semantic change trend, divide the relevant candidate content into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; A third unit, configured to collect interaction data between a user and the content group, where the interaction data includes an interaction type, an interaction duration, an interaction result, and a change in scene state, establish an interaction behavior model based on the interaction data, and calculate the interaction response degree of each content group through the interaction behavior model; A fourth unit, configured to classify content groups according to the interaction response degree, determine a push priority based on the classification result, construct a directed push path graph in combination with the semantic association degree between content groups, where the directed push path graph is used to guide the activation order of content groups, and achieve a smooth transition of the augmented reality experience through the ordered activation of content groups.

[0110] In a third aspect of the embodiments of the present invention, there is provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to call the instructions stored in the memory to execute the method described above.

[0111] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0112] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0113] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. AI-driven interactive AR content intelligent generation and distribution method, characterized in that: include: Acquire scene data collected by the user terminal, and construct a three-dimensional semantic map of the target scene based on the scene data; Calculating semantic entropy values ​​of various regions in the scene according to the three-dimensional semantic map, constructing a scene semantic evolution tree based on the semantic entropy values, and obtaining a semantic change trend of the target scene by updating node states and conversion weights in the scene semantic evolution tree in real time; Calculating the matching degree between the candidate content in the augmented reality content library and the semantic change trend, dividing the candidate content with relevance into content groups according to the matching degree, and generating corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; Collecting interaction data between the user and the content group, the interaction data including interaction type, interaction duration, interaction result, and scene state change, establishing an interaction behavior model based on the interaction data, and calculating the interaction responsiveness of each of the content groups through the interaction behavior model; The content groups are graded according to the interactive responsiveness, and the push priority is determined based on the grading result. A directed push path graph is constructed in combination with the semantic association between the content groups. The directed push path graph is used to guide the activation order of the content groups, and a smooth transition of the augmented reality experience is achieved through the orderly activation of the content groups.

2. The method according to claim 1, characterized in that Calculating the semantic entropy value of each area in the scene according to the three-dimensional semantic map, constructing a scene semantic evolution tree based on the semantic entropy value, and obtaining the semantic change trend of the target scene by updating the node status and conversion weight in the scene semantic evolution tree in real time includes: Dividing the scene space of the three-dimensional semantic map into voxel grids, calculating the probability distribution of the occurrence of each semantic category data in each voxel grid, and calculating the semantic entropy value of each voxel grid based on the probability distribution, wherein the semantic entropy value is used to characterize the semantic complexity of the voxel grid; Acquire a neighborhood voxel set of each voxel grid, calculate the spatial distance between each voxel in the neighborhood voxel set and the target voxel, determine a spatial association weight based on the spatial distance, and perform weighted calculation on the semantic entropy value according to the spatial association weight to obtain a weighted semantic entropy value that takes neighborhood influence into account; Constructing a scene semantic evolution tree based on the weighted semantic entropy value, calculating the semantic state change amount, state transition probability and state transition time interval between adjacent nodes in the scene semantic evolution tree, and using the semantic state change amount, the state transition probability and the state transition time interval as attributes of the connecting edges between the nodes in the scene semantic evolution tree; The similarity between nodes in the scene semantic evolution tree is calculated. The similarity is obtained by a weighted combination of the cosine similarity of the semantic state vector, the Euclidean distance of the entropy distribution matrix, and the time decay function. The scene semantic evolution tree is dynamically updated based on the similarity to obtain the semantic change trend of the scene.

3. The method according to claim 2, characterized in that Constructing a scene semantic evolution tree based on the weighted semantic entropy value, and calculating the semantic state change amount, state transition probability, and state transition time interval between adjacent nodes in the scene semantic evolution tree includes: Extracting a set of semantic attributes of a scene based on the weighted semantic entropy value, wherein the set of semantic attributes includes object category, spatial position, object attribute, and relationship between objects, performing threshold segmentation on the weighted semantic entropy value and extracting semantic regions, obtaining a semantic feature vector for each semantic region, and normalizing the semantic feature vector to obtain an attribute strength value corresponding to each semantic attribute; Constructing a node representation of a scene semantic evolution tree according to the semantic attribute set and the attribute strength value, wherein each node includes a state vector and an entropy value distribution matrix, and constructing the state vector by organizing the attribute strength values ​​into a vector form according to a preset attribute sorting rule, and constructing the entropy value distribution matrix by organizing the weighted semantic entropy values ​​into a matrix form according to a spatial position relationship; Identify adjacent node pairs in the semantic evolution tree, calculate the difference between each component of the state vector between the nodes for each adjacent node pair, divide the difference by the maximum value of the corresponding component and introduce a smoothing factor to obtain a normalized semantic state change; Calculate the binary norm of the semantic state change to obtain the state distance, negate the state distance and perform exponential operation to obtain the initial probability value, and for each node, normalize the initial probability values ​​of all its adjacent nodes to obtain the state transition basic probability; The cosine similarity of the corresponding semantic attributes in the state vector is calculated to obtain the similarity value of each attribute, and the similarity value is weighted and summed with the preset importance weight to obtain the state transition time interval.

4. The method according to claim 1, characterized in that Calculating the matching degree between the candidate content in the augmented reality content library and the semantic change trend, dividing the candidate content with relevance into content groups according to the matching degree, and generating corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map, including: Collecting feature information of candidate content in an augmented reality content library, and converting the feature information into a feature vector, wherein the feature vector includes a visual feature component, a semantic feature component, and an interactive feature component; Performing matching calculation on the feature vector of the candidate content and the semantic change trend, and obtaining a matching score for each candidate content by calculating the consistency between the change direction of the feature vector of the candidate content in the spatial dimension, the attribute dimension and the interaction dimension and the semantic change trend; Taking the vector inner product of the matching scores as the similarity value between the candidate contents, constructing a similarity matrix; calculating the mean and standard deviation of the similarity matrix, setting a dynamic clustering threshold based on the mean and the standard deviation, and classifying the candidate contents with similarity higher than the dynamic clustering threshold into the same content group; Extracting scene features from the three-dimensional semantic map, the scene features including spatial topological structure, semantic attribute distribution and interactive hotspot distribution, and generating corresponding trigger conditions for each content group based on the scene features; Monitoring scene changes in the three-dimensional semantic map, and when the scene changes meet the triggering condition of the content group, calculating an update index of the corresponding content group, the update index combining the access frequency, timeliness and scene relevance of the content group; The update priority of the content group is determined according to the update index, and when the update priority exceeds a preset priority threshold, the content with the highest matching degree with the current semantic change trend is selected from the candidate content library as the update strategy.

5. The method according to claim 1, characterized in that Establishing an interactive behavior model based on the interactive data, and calculating the interactive responsiveness of each content group through the interactive behavior model includes: Extracting interaction time series features, interaction space features, and interaction modal features from the interaction data, and fusing the interaction time series features, the interaction space features, and the interaction modal features to construct a behavior feature vector; An interactive behavior model is established based on the behavior feature vector, wherein the interactive behavior model adopts a hybrid structure of a recursive neural network and an attention mechanism, the temporal dependency of the interactive behavior model is captured by the recursive neural network, the importance of the interactive features is highlighted by the attention mechanism, and a dynamic representation of the interactive behavior model is generated; The interactive responsiveness of the content group is calculated based on the interactive behavior model, the feature vector of the content group is similarly matched with the dynamic representation, and the result of the similarity matching is dynamically weighted by a temporal smoothing factor to obtain a responsiveness score representing the degree of matching between the content group and the interactive behavior.

6. The method according to claim 1, characterized in that Building a directed push path graph in combination with the semantic association between the content groups, wherein the directed push path graph is used to guide the activation sequence of the content groups, and realizing a smooth transition of the augmented reality experience through the orderly activation of the content groups includes: Normalizing the concept weights, entity vectors, and context features in the content group to obtain a standardized feature vector, and constructing a semantic representation of the content group based on the standardized feature vector; Calculating the centrality, density and betweenness centrality of the content group, and obtaining the node importance according to the weighted combination of the centrality, density and betweenness centrality; constructing a directed push path graph based on the node importance, wherein the directed push path graph is used to characterize the activation dependency relationship between the content groups; In the directed push path graph, a conditional probability function considering historical states is used to calculate an activation path score, wherein the conditional probability function takes the semantic association strength, edge weight and resource consumption of adjacent content groups as input parameters, and selects the path with the highest score as the optimal activation path; Preloading the content groups to be activated into the corresponding cache level according to the access order of the optimal activation path, calculating the content differences between adjacent content groups based on the semantic representations of the preloaded content groups, and calculating the context relevance in combination with the real-time scene information; The content difference and the context relevance are input into an adaptive state smoothing function, a transition time between adjacent content groups is dynamically generated, and the activation process of the content group is controlled according to the transition time to achieve a smooth transition of the state.

7. The method according to claim 6, characterized in that According to the access order of the optimal activation path, the content group to be activated is preloaded into the corresponding cache level, the content difference between adjacent content groups is calculated based on the semantic representation of the preloaded content group, and the context relevance is calculated in combination with the real-time scene information, including: Constructing a time series feature vector of the content group, the time series feature vector comprising an access frequency feature, a time interval feature, and a duration feature, and generating a dynamic access pattern of the content group based on the time series feature vector; Calculating a preloading priority score according to the dynamic access mode, the preloading priority score being obtained by a weighted combination of an access frequency score, a time urgency, and a resource requirement, and determining a loading order of the content groups based on the preloading priority score; Perform cache level allocation for the content groups whose loading order is determined, determine the optimal cache level by calculating the difference between the utility value and the storage cost of the content groups at each cache level, load the content groups into the corresponding cache levels, and generate a cache allocation plan; Acquire multimodal features of the content group that has completed cache allocation, perform weighted fusion on the multimodal features to obtain a fusion feature matrix, and generate a semantic embedding vector based on the fusion feature matrix through a nonlinear mapping function; The scene state information, user behavior information and environment information corresponding to the semantic embedding vector are collected, the scene state information, the user behavior information and the environment information are input into a context encoding function to generate a context feature representation, and the context relevance between content groups is calculated based on the context feature representation.

8. An AI-driven interactive AR content intelligent generation and distribution system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain scene data collected by the user terminal and construct a three-dimensional semantic map of the target scene based on the scene data; Calculating semantic entropy values ​​of various regions in the scene according to the three-dimensional semantic map, constructing a scene semantic evolution tree based on the semantic entropy values, and obtaining a semantic change trend of the target scene by updating node states and conversion weights in the scene semantic evolution tree in real time; The second unit is used to calculate the matching degree between the candidate content in the augmented reality content library and the semantic change trend, divide the candidate content with relevance into content groups according to the matching degree, and generate corresponding trigger conditions and update strategies for each content group based on the three-dimensional semantic map; A third unit is used to collect interaction data between the user and the content group, the interaction data including interaction type, interaction duration, interaction result and scene state change, establish an interaction behavior model based on the interaction data, and calculate the interaction responsiveness of each content group through the interaction behavior model; The fourth unit is used to classify the content groups according to the interactive responsiveness, determine the push priority based on the classification results, and construct a directed push path graph in combination with the semantic association between the content groups. The directed push path graph is used to guide the activation order of the content groups, and realize a smooth transition of the augmented reality experience through the orderly activation of the content groups.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images