Intelligent track science popularization content iteration method and system based on immersive interaction

Abstract

The application discloses an intelligent track popular science content iteration method and system based on immersive interaction and belongs to the field of track transportation popular science education information technology. The method comprises the following steps: time alignment and feature concatenation are performed on multi-modal user signals in an immersive interactive environment; a cognitive friction discrimination vector is generated based on cross-domain association of a target understanding state, and a cognitive friction event identifier or a cognitive normal identifier is triggered through threshold comparison; cognitive feedback data of a minimum cognitive unit is inquired in response to a friction event, and a friction response strategy is generated; minimum cognitive units in a single experience are serialized and aggregated into an original cognitive path, path deviation features are extracted by comparing the original cognitive path with a preset path benchmark, and the path deviation features are merged into a group cognitive data set; a group cognitive topology including a consensus area, a transition area, an abnormal area and a dynamic change trend is constructed; and content evolution candidates are derived based on the combination of topology evolution pressure points and cognitive feedback data. The accuracy and content evolution efficiency of the immersive popular science experience are significantly improved.

Description

Technical Field

[0001] This application relates to the field of information technology for popular science education in rail transit, and in particular to a method and system for iterating popular science content in smart rail transit based on immersive interaction. Background Technology

[0002] Currently, with the rapid development of rail transit technology and the increasing demand for improving public scientific literacy, utilizing immersive interactive technologies such as virtual reality and augmented reality to conduct rail transit science education has become an important industry trend. Existing rail transit science education systems typically rely on static graphics, video demonstrations, or simple virtual tours. While these methods improve the presentation to some extent, they still face significant bottlenecks in terms of educational precision and content iteration mechanisms.

[0003] Current content updates primarily rely on expert subjective experience or offline questionnaire feedback, which is time-consuming, costly, and makes it difficult to capture the dynamic evolution of group cognitive structures. The lack of a data-driven mechanism for the aggregation of individual cognitive paths, the construction of group cognitive topologies, and the automatic evolution of content results in popular science content lagging behind the audience's actual cognitive needs for a long time. Summary of the Invention

[0004] To achieve the above objectives, this application provides the following technical solution: According to a first aspect of the present invention, the present invention claims protection for a method for iterating smart rail science popularization content based on immersive interaction, the method comprising the following steps: S1, Time-align the multimodal user signals in the immersive interactive environment and extract basic cognitive state features, including attention representation, load representation and behavioral representation; S2, obtain the target understanding state pre-stored by the system, and perform cross-domain correlation between the attention representation, the load representation and the behavior representation and the target understanding state to generate a cognitive friction discrimination vector. The cognitive friction discrimination vector is used to characterize the degree of deviation between the user's cognitive state and the target understanding state. S3, compare and calibrate the cognitive friction discrimination vector with a preset friction threshold. If the deviation exceeds the preset friction threshold, a cognitive friction event is triggered; if it does not exceed the threshold, it is marked as a cognitive normal event. S4, in response to the cognitive friction event identifier, query the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combine the cognitive friction event identifier and the cognitive feedback data to generate a friction response strategy and update the cognitive feedback data; S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate an original cognitive path. The state outputs include the cognitive friction event identifier or the cognitive normal identifier. A preset path benchmark is obtained from the system. The original cognitive path is compared and calibrated with the preset path benchmark to extract path deviation features. The original cognitive path and its path deviation features are imported into the group cognitive dataset pre-stored in the system. The path deviation features are stored as metadata attributes associated with the corresponding samples. S6, statistically aggregate and modulate the associated stored group cognitive dataset to generate cognitive state distribution features; obtain the system's pre-stored topology construction rules, compare and calibrate the cognitive state distribution features with the topology construction rules, and construct a group cognitive topology, which includes consensus region representation, transition region representation, abnormal region representation, and the dynamic change trend of each region representation; S7. Extract the depth change trend of the transition zone representation, the split state change of the consensus zone representation, and the density change trend of the abnormal zone representation from the group cognitive topology, and jointly determine the evolutionary pressure point; combine the evolutionary pressure point with the cognitive feedback data to generate content evolution candidates.

[0005] Further, in step S1, the multimodal user signals in the immersive interactive environment are time-aligned to extract basic features of cognitive states, including: The multimodal user signals include eye-tracking signals, physiological signals, and interactive operation signals. The multimodal user signals are uniformly mapped to a standardized time axis, and the signals from different sampling frequencies are time-aligned based on an interpolation algorithm to generate a synchronous multimodal signal stream. The eye-tracking signal in the synchronous multimodal signal stream is input into the attention parsing logic to extract the spatial attention distribution representation and the temporal attention persistence representation, and then combined to generate the attention representation. The physiological signals in the synchronous multimodal signal stream are input into the load analysis logic. Based on the skin conductance response variation index and the heart rate variability power ratio index, the cognitive load index and information processing rate are calculated and combined to generate the load characterization. The interactive operation signals in the synchronous multimodal signal stream are input into the behavior parsing logic to extract operation trajectory features and interface dwell features, and then combined to generate the behavior representation. The attention representation, the load representation, and the behavior representation are cascaded to form the basic features of the cognitive state.

[0006] Further, in step S2, the system pre-stores the target understanding state, and cross-domain correlates the attention representation, the load representation, and the behavioral representation with the target understanding state to generate a cognitive friction discriminant vector, including: The attention representation, the load representation, and the behavior representation are mapped to the same multidimensional feature space to construct a three-dimensional cognitive state tensor. The target understanding state vector is obtained from the system’s pre-stored target understanding state vector, which is generated by the expert-annotated understanding standard corresponding to the smallest cognitive unit or the feature center of historical high-understanding users. Calculate the spatial deviation between the three-dimensional cognitive state tensor and the target understanding state vector, and encode the spatial deviation and its directional components into the cognitive friction discrimination vector. The magnitude of the cognitive friction discrimination vector represents the degree of deviation, and the directional components represent the deviation dimension.

[0007] Further, in step S3, the cognitive friction discrimination vector is compared and calibrated with a preset friction threshold, including: The preset friction threshold is obtained, which includes a global baseline threshold and an individual adaptive threshold. The individual adaptive threshold is dynamically determined based on the statistical quantile of the distribution of the user's historical cognitive friction events or the upper limit of the confidence interval of the user's baseline cognitive state. Calculate the magnitude of the cognitive friction discrimination vector; If the modulus value exceeds the global baseline threshold and the individual adaptive threshold, the cognitive friction event identifier is triggered, and the friction type label is recorded. The friction type label includes attention deficit type, overload type, and behavioral deviation type. If the modulus value does not exceed both the global baseline threshold and the individual adaptive threshold simultaneously, it is marked as the cognitively normal identifier.

[0008] Further, in step S4, in response to the cognitive friction event identifier, the system queries the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combines the cognitive friction event identifier with the cognitive feedback data to derive a friction response strategy and update the cognitive feedback data, including: The smallest cognitive unit currently presented is determined. The smallest cognitive unit is the smallest presentation granularity in the orbital science popularization content that has independent knowledge semantics and cannot be further divided. Each smallest cognitive unit has a unique unit identifier. The system queries the pre-stored cognitive feedback database based on the unique unit identifier to obtain the cognitive feedback data corresponding to the smallest cognitive unit. The cognitive feedback data includes expected understanding time, a common misunderstanding pattern library, a multimodal auxiliary resource index, and historical friction response records. Based on the friction type label in the cognitive friction event identifier, the corresponding misunderstanding pattern in the common misunderstanding pattern library is matched, and the auxiliary explanation resources associated in the multimodal auxiliary resource index are called to generate the friction response strategy; The cognitive friction event identifier, the friction response strategy, and the response time are associated and recorded to generate a friction response record; The friction response records are merged into the historical friction response records of the smallest cognitive unit, and the statistical features in the cognitive feedback data are updated using a sliding time window.

[0009] Further, in step S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate the original cognitive path, including: Extract the unique unit identifier and its corresponding state output of each smallest cognitive unit presented sequentially in a single experience according to the time sequence, and construct a time-ordered state sequence. The state output is the cognitive friction event identifier or the cognitive normal identifier. The state sequence is encoded into a directed graph structure, where each node is a unique unit identifier of the smallest cognitive unit, the node attribute is the corresponding state output, and the edge is the transition relationship between adjacent smallest cognitive units, thereby generating the original cognitive path; Obtain the preset path benchmark stored in the system. The preset path benchmark is generated by the standard understanding path marked by domain experts or the cognitive path cluster center of historical high-performance user groups. The original cognitive path is aligned with the preset path benchmark by nodes and sequence comparison, and path deviation features are extracted. The path deviation features include node skip rate, node backtracking rate, path detour degree, friction node density, and deviation degree of the cognitive state at the end point. The original cognitive path and its path deviation features are imported into the pre-stored group cognitive dataset of the system, and the path deviation features are stored as metadata attributes of the sample and associated with the original cognitive path.

[0010] Further, in step S6, the associated and stored group cognitive dataset is statistically aggregated and subjected to regional weight modulation to generate cognitive state distribution features and construct a group cognitive topology, including: Kernel density estimation was performed on samples in the aforementioned group cognitive dataset according to cognitive state characteristics, and the distribution of friction frequency, load mean, and attention concentration in each local region was statistically analyzed. The samples are subjected to regional weight modulation, and the regional weights are multiplied and weighted according to the timeliness weight factor and the user experience level weight factor of the samples, wherein the timeliness weight factor decays as the interval between the sample collection time and the current time increases. The weighted samples are mapped to the cognitive state feature space, graph node connection relationships are constructed based on the cognitive state similarity between samples, and the region boundaries are divided based on the local density of the samples to generate the group cognitive topology. The consensus region is characterized as a set of samples in a high-density connected subgraph where the friction frequency is lower than the consensus threshold and the average load is in the middle range. The transition region is characterized as a set of boundary samples connecting different high-density connected subgraphs, with a local density lower than that of the consensus region and higher than that of the anomaly region. The abnormal region is characterized as a set of samples with a local density lower than the abnormal density threshold or a friction frequency higher than the abnormal friction threshold. The dynamic change trends of each region are obtained by differential calculation of the regional boundary displacement, sub-cluster number change and sample density change of the group cognitive topology under different time windows.

[0011] Further, in step S7, evolutionary pressure points are extracted from the group cognitive topology and combined with the cognitive feedback data to generate content evolution candidates, including: Extract the depth change trend of the transition region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When the rate of migration of the transition region sample to the abnormal region exceeds the first preset rate threshold, it is determined that there is transition region evolution pressure. Extract the splitting state changes of the consensus region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When multiple unconnected subclusters appear inside the consensus region and the center distance between the subclusters exceeds a preset splitting distance threshold, it is determined that there is consensus region splitting evolution pressure. Extract the density change trend of the abnormal region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When the growth rate of the abnormal region sample density exceeds the second preset rate threshold, it is determined that there is pressure for the expansion and evolution of the abnormal region. When at least two of the following conditions are met simultaneously: the transition zone evolution pressure, the consensus zone splitting evolution pressure, and the abnormal zone expansion evolution pressure, the corresponding smallest cognitive unit is marked as the evolution pressure point. Extract cognitive feedback data of the smallest cognitive unit associated with the evolutionary pressure point, analyze the historical friction response records and common misunderstanding patterns in the cognitive feedback data, and generate content evolution candidates. The content evolution candidates include: the smallest cognitive unit presentation order adjustment scheme, the supplementary explanation node insertion position, the interaction difficulty gradient reconstruction scheme, and the multimodal auxiliary resource replacement strategy.

[0012] Furthermore, following step S7, step S8 is also included: The content evolution candidates were verified by A / B testing in the test user group, and the occurrence rate of cognitive friction event identifiers and the change in path deviation characteristics of the test users were collected. If the occurrence rate of the cognitive friction event identifier decreases and the change in the path deviation feature converges to the preset convergence interval, the verified content evolution candidate will be deployed as a new track science popularization content presentation scheme, and the cognitive feedback database and the preset path benchmark will be updated simultaneously.

[0013] According to a second aspect of the present invention, the present invention claims protection for an iterative system for intelligent rail science popularization content based on immersive interaction, comprising: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, enable the processors to implement the immersive interactive smart rail science popularization content iteration method.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The beneficial effects of the technical solutions provided by the embodiments of the present invention include: By temporally aligning and concatenating features from multimodal user signals in an immersive interactive environment, basic cognitive state features, including attention representation, workload representation, and behavioral representation, are extracted. Based on the target understanding state, cross-domain association is performed to generate a cognitive friction discrimination vector, which triggers a cognitive friction event identifier or a cognitive normality identifier through threshold comparison. In response to friction events, cognitive feedback data of the smallest cognitive unit is queried, and a friction response strategy is generated. The smallest cognitive unit in a single experience is serialized and aggregated into an original cognitive path, compared with a preset path benchmark to extract path deviation features, and then incorporated into a group cognitive dataset. Statistical aggregation and regional weight modulation are performed on the group cognitive dataset to construct a group cognitive topology including consensus zones, transition zones, abnormal zones, and dynamic changing trends. Content evolution candidates are derived based on the combination of topological evolution pressure points and cognitive feedback data. This invention significantly improves the accuracy and content evolution efficiency of immersive science popularization experiences. Attached Figure Description

[0015] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Figure 1 This is a schematic diagram illustrating the implementation process of the intelligent rail science popularization content iteration method based on immersive interaction provided in the embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0017] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0018] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] According to the first embodiment of the present invention, referring to Figure 1 This invention seeks to protect a method for iterating smart rail science popularization content based on immersive interaction, the method comprising the following steps: S1, Time-align the multimodal user signals in the immersive interactive environment and extract basic cognitive state features, including attention representation, load representation and behavioral representation; S2, obtain the target understanding state pre-stored by the system, and perform cross-domain correlation between the attention representation, the load representation and the behavior representation and the target understanding state to generate a cognitive friction discrimination vector. The cognitive friction discrimination vector is used to characterize the degree of deviation between the user's cognitive state and the target understanding state. S3, compare and calibrate the cognitive friction discrimination vector with a preset friction threshold. If the deviation exceeds the preset friction threshold, a cognitive friction event is triggered; if it does not exceed the threshold, it is marked as a cognitive normal event. S4, in response to the cognitive friction event identifier, query the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combine the cognitive friction event identifier and the cognitive feedback data to generate a friction response strategy and update the cognitive feedback data; S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate an original cognitive path. The state outputs include the cognitive friction event identifier or the cognitive normal identifier. A preset path benchmark is obtained from the system. The original cognitive path is compared and calibrated with the preset path benchmark to extract path deviation features. The original cognitive path and its path deviation features are imported into the group cognitive dataset pre-stored in the system. The path deviation features are stored as metadata attributes associated with the corresponding samples. S6, statistically aggregate and modulate the associated stored group cognitive dataset to generate cognitive state distribution features; obtain the system's pre-stored topology construction rules, compare and calibrate the cognitive state distribution features with the topology construction rules, and construct a group cognitive topology, which includes consensus region representation, transition region representation, abnormal region representation, and the dynamic change trend of each region representation; S7. Extract the depth change trend of the transition zone representation, the split state change of the consensus zone representation, and the density change trend of the abnormal zone representation from the group cognitive topology, and jointly determine the evolutionary pressure point; combine the evolutionary pressure point with the cognitive feedback data to generate content evolution candidates.

[0020] This embodiment is presented in the full picture of its application in an immersive digital twin science popularization experience hall for rail transit.

[0021] Participants enter an immersive interactive environment, wearing head-mounted displays, wearable physiological sensor arrays, and interactive handles to engage in a science popularization experience on the theme of high-speed railway traction drive systems. The system breaks down the knowledge system of this topic into five smallest cognitive units: traction substation power conversion, pantograph current collection, main transformer voltage conversion, traction converter power conversion, and traction motor drive output. Each smallest cognitive unit has a unique unit identifier and is an indivisible semantic granularity of knowledge.

[0022] Once the user enters the virtual scenario of traction substation power conversion, the system begins execution S1. The head-mounted display terminal collects the user's eye movement signals at a preset frequency, including gaze point coordinates and pupil diameter; the wearable physiological sensor array collects skin conductance response signals and heart rate variability signals at a preset frequency; and the interactive controller collects operation trajectories and click events at a preset frequency. Because the sampling frequencies of the three signals are different, the system first maps them uniformly to a standardized time axis, uses an interpolation algorithm for time alignment, and generates a synchronous multimodal signal stream.

[0023] Subsequently, the system inputs eye-tracking signals from the synchronous multimodal signal stream into the attention analysis logic to extract the spatial distribution of the user's attention to key components such as the high-voltage circuit breaker and transformer windings in the virtual scene, as well as the temporal continuity of attention to each component, and combines them to generate an attention representation. Physiological signals are input into the load analysis logic, and based on skin conductance response variability indicators and the low-frequency / high-frequency power ratio of heart rate variability, the system calculates the user's current cognitive load index and information processing rate, and combines them to generate a load representation. Interactive operation signals are input into the behavior analysis logic to extract the user's controller operation trajectory features and interface dwell features on interactive menus in the scene, and combine them to generate a behavior representation. Finally, the system concatenates the attention representation, load representation, and behavior representation to form the basic features of the user's cognitive state at the current moment.

[0024] The system executes S2. For the smallest cognitive unit of power conversion in the traction substation, the system's pre-stored target understanding state vector, annotated by experts in the rail transit field, represents the typical attention distribution, load level, and interaction pattern of a high-understanding user in this scenario. The system maps the user's basic cognitive state features and the target understanding state vector to the same multi-dimensional feature space, constructing a three-dimensional cognitive state tensor and calculating the spatial deviation between them. This spatial deviation and its directional component are encoded as a cognitive friction discriminant vector. Its magnitude quantifies the degree of deviation between the user's current cognitive state and the target understanding state, while the directional component indicates whether the deviation mainly occurs in the attention, load, or behavioral dimensions.

[0025] The system executes step S3. The system obtains preset friction thresholds, where the global baseline threshold is set by the system based on large-scale user data statistics, and the individual adaptive threshold is dynamically determined based on the statistical quantiles of the distribution of cognitive friction events in the user's historical experiences. The system calculates the magnitude of the cognitive friction discrimination vector and finds that this magnitude exceeds both the global baseline threshold and the individual adaptive threshold, and the directional component shows a significant deviation in the load dimension. Therefore, a cognitive friction event is triggered, and the friction type label is recorded as overload type.

[0026] The system executes S4. In response to the cognitive friction event identifier, the system determines the smallest cognitive unit currently presented is the traction substation power conversion and queries the cognitive feedback database based on its unique unit identifier. The cognitive feedback data corresponding to this unit in the database includes: expected understanding time, a common misunderstanding pattern library, a multimodal auxiliary resource index, and historical friction response records. Based on the overload type tag, the system matches the corresponding pattern in the common misunderstanding pattern library that leads to excessive load due to conceptual abstraction, calls the 3D current path animation in the multimodal auxiliary resource index as an auxiliary explanation resource, and generates a friction response strategy: pausing the current high-density information output, inserting a 3D animation demonstration, and providing a degraded voice explanation. The system associates and records the cognitive friction event identifier, friction response strategy, and response time, generating a friction response record, and uses a sliding time window to update the historical friction response records and statistical characteristics of this unit.

[0027] The user continued to experience subsequent units. In the pantograph current collection unit, the user's cognitive state matched the target understanding state, and the system marked the cognitive state as normal. In the main transformer voltage transformation unit, due to difficulty in understanding the main transformer oil circulation cooling mechanism, the cognitive friction event was triggered again, and the system called local magnification and hotspot labeling resources to respond. In the remaining units, the user was marked as having normal cognitive state.

[0028] The system executes S5. It extracts the identifiers and state outputs of each smallest cognitive unit presented sequentially during a single experience, constructing a time-ordered state sequence. The system encodes this state sequence into a directed graph structure, where nodes are unique identifiers of each smallest cognitive unit, node attributes are the corresponding state outputs, and edges represent the transition relationships between adjacent units, generating the experiencer's original cognitive path.

[0029] The system acquires a preset path benchmark, which is jointly generated by standard understanding paths annotated by domain experts and cognitive path cluster centers of historical high-performing user groups. The system performs node alignment and sequence comparison between the original cognitive paths and the preset path benchmark, extracting path deviation features: node skip rate, node backtracking rate, path detour degree, friction node density, and deviation from the endpoint cognitive state. The system integrates the original cognitive paths and their path deviation features into a group cognitive dataset, storing the path deviation features as metadata attributes associated with the original cognitive paths.

[0030] The system executes S6. Kernel density estimation is performed on recent data in the group cognitive dataset based on cognitive state characteristics. The distribution of friction frequency, load mean, and attention concentration in each local region is statistically analyzed. The system modulates regional weights for the samples. The timeliness weight factor decreases according to a preset decay law as the interval between the sample collection time and the current time increases. The user experience level weight factor is determined based on the user's self-assessed rail transit knowledge background level. The product of these two factors constitutes the regional weight. The weighted samples are mapped to the cognitive state feature space. Graph node connections are constructed based on the similarity of cognitive states between samples. Regional boundaries are delineated based on local density, generating a group cognitive topology.

[0031] In this topology, the region where the pantograph current collection and traction motor drive output are located forms a high-density connected subgraph. The friction frequency is below the consensus threshold and the average load is in the medium range, thus it is characterized as the consensus region. The traction converter's power conversion is located at the boundary connecting the consensus region and the abnormal region, with a local density between the two, thus it is characterized as the transition region. The region where the traction substation's power conversion and the main transformer's voltage conversion are located has a low local density and a high friction frequency, thus it is characterized as the abnormal region. The system obtains the dynamic change trend of each region's characterization by performing differential calculations on the regional boundary displacement, sub-cluster number changes, and sample density changes under different time windows.

[0032] The system executes S7. From the dynamic change trend of the group's cognitive topology, it extracts the depth change trend of the transition zone representation. It finds that the rate at which the traction converter's power conversion samples migrate from the transition zone to the abnormal zone exceeds a first preset rate threshold, indicating the existence of transition zone evolution pressure. It also extracts the splitting state changes of the consensus zone representation, finding that although the pantograph current collection and traction motor drive output belong to the same consensus zone, two unconnected subclusters exist within them, with the center distance between the subclusters exceeding a preset splitting distance threshold, indicating the existence of consensus zone splitting evolution pressure. Finally, it extracts the density change trend of the abnormal zone representation, finding that the abnormal zone sample density growth rate of the traction substation's power conversion and the main transformer's voltage conversion exceeds a second preset rate threshold, indicating the existence of abnormal zone expansion evolution pressure. Since both transition zone evolution pressure and abnormal zone expansion evolution pressure are satisfied simultaneously, the system marks the traction substation's power conversion and the main transformer's voltage conversion as evolution pressure points.

[0033] The system extracts cognitive feedback data from the smallest cognitive units associated with the evolving pressure points, analyzes historical friction response records and common misunderstanding patterns, and generates content evolution candidates. For traction substation power transformation, it proposes a presentation order adjustment scheme, moving the unit from the beginning to after pantograph current collection, establishing perceptual cognition before learning abstract concepts. It also proposes a supplementary explanation node insertion position, inserting a power grid to train energy flow overview node before the unit. Furthermore, it proposes an interactive difficulty gradient reconstruction scheme to reduce the disassembly steps of high-voltage equipment. Finally, it proposes a multimodal auxiliary resource replacement strategy, replacing textual descriptions with interactive 3D disassembly models. Similar content evolution candidates are proposed for main transformer voltage transformation.

[0034] Furthermore, the specific implementation of multimodal signal temporal alignment and cognitive state basic feature extraction in step S1 is as follows: The signal is uniformly mapped to a standardized time axis, and the time alignment of signals from different sampling frequencies is performed based on an interpolation algorithm to generate a synchronous multimodal signal stream; Specifically, the head-mounted display terminal outputs a raw sampling rate of f. E =60Hz gaze point coordinate stream (x t ,y t ) and pupil diameter flow d t , (x t ,y t Normalized to screen space d t The unit is millimeters; the raw sampling rate of the wearable physiological sensor array output is f. P =32Hz skin conductance response signal EDA(t) and RR interval sequence RR(t), EDA(t) is in microsiemens and RR(t) is in milliseconds; the original sampling rate of the interactive handle output is f I =30Hz Handheld Space Trajectory Flow (x top ,y top ) and interactive event stream e t , (x top ,y top The unit is meters, e t Encode the event type.

[0035] The system uses the master clock of the immersive content rendering engine as a reference to set the standardized timeline target sampling frequency. For values ​​below f T For any original signal sequence X={x(t)}, upsampling alignment is performed using a cubic spline interpolation algorithm: i )}, on the target time axis {τ j The value on} is , where t k ≤τ j ≤tk+1 a k b k c k d k These are spline coefficients, determined by the natural boundary conditions. Confirmed, S ’’ The second derivatives t1 and t of the spline function N These represent the beginning and end times of the signal, respectively; for values ​​higher than f... T The signal stream is aligned using mean downsampling. For continuous missing segments in the eye-tracking data caused by blinking, linear interpolation is used to fill in the gaps, taking the arithmetic mean of the valid data before and after the missing segment. The aligned synchronous multimodal signal stream is denoted as S={E(t),P(t),I(t)}, where t is the time index, t∈{1,2,...,T}, and T is the total number of frames in a single experience, and is fed into the parallel parsing pipeline.

[0036] The eye-tracking signal in the synchronous multimodal signal stream is input into the attention parsing logic to extract the spatial attention distribution representation and the temporal attention persistence representation, and then combined to generate the attention representation. Specifically, the attention parsing logic is deployed in the eye-tracking signal pipeline, first processing the gaze coordinate stream (x... t ,y t Two-dimensional Gaussian kernel density estimation is performed to generate a spatial interest distribution representation on a 32×32 grid. The kernel function is Where u and v are the differences between the grid center and the gaze point coordinates, h x h y For bandwidth parameters, select according to Silverman rules. σ is the standard deviation of the gaze point coordinates within the window, and N is the total number of valid gaze points within the window; secondly, the pupil diameter flow d t Perform temporal differentiation and entropy calculations to calculate the sample entropy of the pupil diameter sequence (embedding dimension m=2, similarity tolerance r=0.2×std(d)). t std(d) t The sum of the standard deviation of the pupil diameter sequence and the first three moments (mean, variance, skewness) of the fixation duration distribution is used to construct a temporal representation of attention duration. The two are combined to generate a complete attention representation A=[vec(A spatial A temporal ], vec(∙) is a matrix vectorization operator with a total dimension of 1028.

[0037] The physiological signals in the synchronous multimodal signal stream are input into the load analysis logic. Based on the skin conductance response variation index and the heart rate variability power ratio index, the cognitive load index and information processing rate are calculated and combined to generate the load characterization. Specifically, the load parsing logic is deployed in the physiological signal pipeline to extract the phase component of the skin conductance response signal EDA(t) using a 0.05~5Hz bandpass filter. Local maxima exceeding the baseline by 0.01μS are detected as peak values, and the variation index of skin conductance response within the statistical window is calculated. , Let N be the amplitude of the i-th peak. SCR μ represents the total number of peak values. SCR and σ SCR The historical mean and standard deviation of the skin conductance variability index for this user at rest were calculated. The power spectral density of the RR interval sequence RR(t) was calculated by spline interpolation and resampling to 4Hz, followed by Fast Fourier Transform. The low-frequency band LF = 0.04~0.15Hz and the high-frequency band HF = 0.15~0.40Hz were defined. The power ratio of heart rate variability was calculated. P LF P HF These are the integrated power values ​​for the LF and HF bands, respectively. The cognitive load index is calculated by combining these two values. μ HRV and σ HRV The historical mean and standard deviation of the power ratio of heart rate variability for this user at rest are given, tanh(∙) is the hyperbolic tangent function, and weighting coefficients of 0.6 and 0.4 are empirical calibration values; the information processing rate is calculated. T win =5, in seconds, represents the length of the sliding window. Mutual information operator; the above index combination generates load characterization. .

[0038] The interactive operation signals in the synchronous multimodal signal stream are input into the behavior parsing logic to extract operation trajectory features and interface dwell features, and then combined to generate the behavior representation. Specifically, the behavior analysis logic is deployed in the interactive operation signal pipeline to perform curvature and dwell point clustering analysis on the operation trajectory: calculating the total path length of the operation trajectory within the window. , Euclidean norm; mean curvature Number of stops N stay (A spatial point with a velocity below 0.01 m / s and a duration exceeding 500 ms is counted as one dwell point); these constitute the characteristics of the operational trajectory. The system performs time interval statistics and state machine analysis on the interaction event flow, and calculates the percentage of cumulative dwell time on each UI element. j=1,...,J, take the maximum residency percentage and interface switching frequency f switch , forming interface dwell characteristics The combination of the two generates a behavioral representation. .

[0039] The attention representation, the load representation, and the behavior representation are cascaded to form the basic features of the cognitive state.

[0040] Specifically, the feature concatenation module performs Z-score standardization on the above three sets of features to unify the dimensions, and concatenates them into a unified vector according to a preset dimensional order to form the basic features of cognitive state. The total dimension is 1035 = 1028 + 2 + 5, and is output to the subsequent step S2.

[0041] Further, in step S2, the system pre-stores the target understanding state, and cross-domain correlates the attention representation, the load representation, and the behavioral representation with the target understanding state to generate a cognitive friction discriminant vector, including: The attention representation, the load representation, and the behavior representation are mapped to the same multidimensional feature space to construct a three-dimensional cognitive state tensor.

[0042] Specifically, the system splits the basic cognitive state features F from step S1 into modalities. Through pre-trained linear projection matrices Mapping to the shared 64-dimensional latent space, we obtain The three are then stacked into a three-dimensional cognitive state tensor. Its elements This outer product operation is used to capture the nonlinear interaction between modes.

[0043] The system obtains a pre-stored target understanding state vector, which is generated by the expert-annotated understanding standard corresponding to the smallest cognitive unit or the feature center of a historical high-understanding user.

[0044] Specifically, the system queries the pre-stored target understanding state vector G based on the unique identifier uid of the current smallest cognitive unit. u The vector is generated using one of two methods: Method 1, where five domain experts experience the unit under standard understanding conditions, collect tensor features, and then take the arithmetic mean. , Let G be the cognitive state tensor of the m-th expert; Method 2: Select user samples from the group cognitive dataset whose cognitive friction event incidence rate is less than 5% and whose path deviation features are in the top 10%, perform K-means clustering (K=1) on their tensor features, and take the cluster center as G. u .

[0045] Calculate the spatial deviation between the three-dimensional cognitive state tensor and the target understanding state vector, and encode the spatial deviation and its directional components into the cognitive friction discrimination vector. The magnitude of the cognitive friction discrimination vector represents the degree of deviation, and the directional components represent the deviation dimension.

[0046] Specifically, the system uses a weighted cosine-Euclidean hybrid distance to calculate the spatial deviation. , The Frobenius inner product of tensors It is the Frobenius norm. The direction components are also calculated. , The target state G is respectively u Projection on the three modes The deviation and direction components are encoded into a cognitive friction discrimination vector. Its modulus The total deviation is represented by the directional component, which represents the deviation dimension, and is output to the downstream step S3.

[0047] Further, in step S3, the cognitive friction discrimination vector is compared and calibrated with a preset friction threshold, including: The preset friction threshold is obtained, which includes a global baseline threshold and an individual adaptive threshold. The individual adaptive threshold is dynamically determined based on the statistical quantile of the distribution of the user's historical cognitive friction events or the upper limit of the confidence interval of the user's baseline cognitive state.

[0048] Specifically, the system obtains the global baseline threshold. These represent the mean and standard deviation δ of the deviation from the cognitively normal samples in the historical database, respectively. An individual adaptive threshold is calculated based on the friction records of the user's historical H experiences. , The function is the 85th percentile. When H < 10, the upper limit of the baseline confidence interval is used. This is a statistical measure of the user's deviation in the resting state during the first 3 minutes of the experience.

[0049] Calculate the magnitude of the cognitive friction discrimination vector.

[0050] Specifically, the system uses cognitive friction discrimination vectors The first dimension component δ is extracted as the modulus.

[0051] If the modulus value exceeds both the global baseline threshold and the individual adaptive threshold, the cognitive friction event identifier is triggered, and a friction type label is recorded. The friction type label includes inattention-deficient, overload-type, and behavioral deviation-type. If the modulus value does not exceed both the global baseline threshold and the individual adaptive threshold simultaneously, it is marked as the cognitively normal identifier.

[0052] Specifically, the system determines when When both conditions are met, a cognitive friction event flag (Flag=1) is triggered; otherwise, it is marked as Flag=0. When Flag=1, the system compares the direction components to record the friction type label: when... When it is determined to be an overload type, when It was diagnosed as attention deficit type; when If the three criteria are not significantly different, the type is classified as behavioral deviation; 1.2 is the preset margin coefficient. If none of the three criteria are significantly different, the type is classified as mixed, and the largest of the three criteria is taken as the main type.

[0053] Further, in step S4, in response to the cognitive friction event identifier, the system queries the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combines the cognitive friction event identifier with the cognitive feedback data to derive a friction response strategy and update the cognitive feedback data, including: The smallest cognitive unit currently presented is determined. The smallest cognitive unit is the smallest presentation granularity in the orbital science popularization content that has independent knowledge semantics and cannot be further divided. Each smallest cognitive unit has a unique unit identifier.

[0054] Specifically, the system reads the content module being presented from the rendering state of the current immersive scene and determines the unique identifier (uid) of its corresponding smallest cognitive unit.

[0055] The system queries the pre-stored cognitive feedback database based on the unique unit identifier to obtain the cognitive feedback data corresponding to the smallest cognitive unit. The cognitive feedback data includes the expected understanding time, a library of common misunderstanding patterns, a multimodal auxiliary resource index, and historical friction response records.

[0056] Specifically, the system queries the cognitive feedback database based on the uid to obtain the cognitive feedback data for that unit, including: expected comprehension time T. exp The library includes a common misunderstanding pattern library (MisUndLib, each record contains a pattern number, a list of keywords, and a 128-dimensional semantic embedding vector), a multimodal auxiliary resource index (ResIndex, each record contains a resource number, resource type, storage address, and tag vector), and historical friction response records (HistResp).

[0057] Based on the friction type label in the cognitive friction event identifier, the corresponding misunderstanding pattern in the common misunderstanding pattern library is matched, and the auxiliary explanation resources associated in the multimodal auxiliary resource index are invoked to generate the friction response strategy.

[0058] Specifically, the system performs a dual-path retrieval: first, a hard match is performed using the friction type label Type, and then the current cognitive state tensor T is processed by the encoder E. mis Mapped to query vector Calculate the cosine similarity with the embedding vectors of each entry in MisUndLib. Given the embedding vector of the j-th misunderstanding pattern, recall The TOP-3 pattern is then used. The tag matching score in ResIndex is then calculated. Auxiliary resources greater than 0.5 The set of tags for the current recall misunderstanding pattern, tags r For the list of tags of the r-th resource This is the indicator function. The system follows the policy utility function. Generate the optimal friction response strategy, where match(s,Type) is the degree of matching between the strategy template and the current friction type. The historical efficiency of this strategy on this unit.

[0059] The cognitive friction event identifier, the friction response strategy, and the response time are associated and recorded to generate a friction response record.

[0060] Specifically, the system associates the current cognitive friction event identifier Flag, friction type label Type, generated friction response strategy Strategy, and the timestamp t of the response time to generate a friction response record R=(t,Flag,Type,Strategy).

[0061] The friction response records are merged into the historical friction response records of the smallest cognitive unit, and the statistical features in the cognitive feedback data are updated using a sliding time window.

[0062] Specifically, after the strategy is executed, the system calculates the validity of the strategy. , and These represent the number of friction events in the first three units of strategy execution and the total number of units, respectively. and N after The system records the number of friction events in the three units following strategy execution and the total number of units. The effects are then added to the friction response record and merged into the historical friction response record of the smallest cognitive unit. The system uses a 30-day sliding time window to update statistical features. H represents the total number of historical records, with only the most recent 30 records retained for calculation.

[0063] Further, in step S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate the original cognitive path, including: Extract the unique unit identifier and its corresponding state output of each smallest cognitive unit presented sequentially in a single experience according to time order, and construct a time-ordered state sequence. The state output is the cognitive friction event identifier or the cognitive normality identifier.

[0064] Specifically, the system extracts the identifiers (uid) of the smallest cognitive units presented sequentially in a single experience, according to time sequence. t and the corresponding status output Flags are used to identify cognitive friction events. t =0 indicates normal cognitive function; a time-ordered state sequence is constructed. , where t is the time sequence index and N is the total number of the smallest cognitive units presented in a single experience.

[0065] The state sequence is encoded into a directed graph structure, where each node is a unique identifier of the smallest cognitive unit, the node attribute is the corresponding state output, and the edge is the transition relationship between adjacent smallest cognitive units, thus generating the original cognitive path.

[0066] Specifically, the system encodes the state sequence into a directed graph. Node attributes include , Duration of stay edge weight From Transferred to The transfer time is used to generate the original cognitive path.

[0067] Obtain the preset path benchmark stored in the system. The preset path benchmark is generated by the standard understanding path marked by domain experts or the cognitive path cluster center of historical high-performance user groups.

[0068] Specifically, the system obtains a preset path baseline. The graph is generated by fusing the results of two methods: Method 1, where five domain experts each annotate the standard understanding path, and the majority vote is used as the baseline graph structure; Method 2, high-performing users (cognitive friction event rate below 5% and endpoint test accuracy above 90%) are selected from the group cognitive dataset, and their paths are clustered using Levenshtein distance, with the cluster center of the largest cluster used as the baseline. The results of the two methods are then weighted and fused. .

[0069] The original cognitive path is aligned with the preset path benchmark by nodes and sequence comparison, and path deviation features are extracted. The path deviation features include node skip rate, node backtracking rate, path detour degree, friction node density, and deviation degree of the cognitive state at the destination.

[0070] Specifically, the system uses dynamic time warping for node alignment: Among them, the measure of difference like The value is 0 when the paths are identical and 1 when they are different. The five-dimensional path deviation feature is then calculated after alignment. Node skip rate , For the set of baseline path nodes, The actual set of path nodes; node backoff rate. Path detour , , Shortest time for the baseline path; friction node density Deviation from endpoint cognitive state , For the cognitive state tensor of the last unit at the end of the experience, This represents the target understanding state for this unit.

[0071] The original cognitive path and its path deviation features are imported into the pre-stored group cognitive dataset of the system, and the path deviation features are stored as metadata attributes of the sample and associated with the original cognitive path.

[0072] Specifically, the system will use the original cognitive path and its path deviation characteristics This sample was incorporated into the group cognition dataset. The sample structure is , The metadata is associated and stored in the sample header.

[0073] Further, in step S6, the associated and stored group cognitive dataset is statistically aggregated and subjected to regional weight modulation to generate cognitive state distribution features and construct a group cognitive topology, including: Kernel density estimation is performed on the samples in the group cognitive dataset according to cognitive state characteristics, and the friction frequency distribution, load mean distribution and attention concentration distribution of each local region are statistically analyzed.

[0074] Specifically, the system uses a group cognition dataset. Two-dimensional kernel density estimation of samples based on cognitive state characteristics , The total number of samples in the dataset. For the first Two feature values ​​of a sample, , Two-dimensional bandwidth, adaptive bandwidth , For the first The sample to its number The distance between neighbors The distribution of friction frequency, average load, and attention concentration in each local area was statistically analyzed.

[0075] The samples are subjected to regional weight modulation, and the regional weights are multiplied and weighted according to the timeliness weight factor and the user experience level weight factor of the samples, wherein the timeliness weight factor decays as the interval between the sample collection time and the current time increases.

[0076] Specifically, the system performs weight modulation on the samples. Timeliness weight , The current system time. The sample collection time is 90, and the decay constant is 90; the empirical level weight is 90. The weighting is set at 1.0 for the general public, 1.2 for professional students, and 1.5 for practitioners. The level is determined by the user's self-assessment of their rail transit knowledge background filled in during registration.

[0077] The weighted samples are mapped to the cognitive state feature space, graph node connections are constructed based on the cognitive state similarity between samples, and regional boundaries are divided based on the local density of samples to generate the group cognitive topology.

[0078] Specifically, the system maps the weighted samples to the cognitive state feature space and calculates the cosine similarity between samples. ,when Establish connections and construct a k-nearest neighbor graph. Then, use the HDBSCAN algorithm to divide the region, with the minimum cluster size set to 10 and the minimum number of samples set to 5.

[0079] The consensus region is characterized as a set of samples in a high-density connected subgraph where the friction frequency is below the consensus threshold and the average load is in the middle range.

[0080] Specifically, the consensus area is and A high-density connected subgraph.

[0081] The transition region is characterized as a set of boundary samples connecting different high-density connected subgraphs, with a local density lower than that of the consensus region and higher than that of the anomaly region.

[0082] Specifically, the transition zone is a local density. Boundary samples, This represents the normalized local density output by the HDBSCAN algorithm.

[0083] The abnormal region is characterized as a set of samples where the local density is lower than the abnormal density threshold or the friction frequency is higher than the abnormal friction threshold.

[0084] Specifically, the abnormal region is a local density or friction frequency The sample set.

[0085] The dynamic change trends of each region are obtained by differential calculation of the regional boundary displacement, sub-cluster number change and sample density change of the group cognitive topology under different time windows.

[0086] Specifically, the system maintains three time windows of 7 days, 30 days, and 90 days, and calculates the regional boundary displacement through differential calculation. , For the first Centroid coordinates of the region within a window; changes in the number of subclusters. Sample density change , for Normalized sample density of the region within a window.

[0087] Further, in step S7, evolutionary pressure points are extracted from the group cognitive topology and combined with the cognitive feedback data to generate content evolution candidates, including: The depth change trend of the transition region representation is extracted from the dynamic change trend of the representation of each region in the group cognitive topology. When the migration rate of the transition region sample to the abnormal region exceeds the first preset rate threshold, it is determined that there is evolutionary pressure in the transition region.

[0088] Specifically, the system sets a first preset rate threshold. Samples / day. Extracting the deep change trend of the transition zone from the dynamic change trend, as the migration rate from the transition zone to the anomaly zone... hour, For time window The number of samples migrating from the transition zone to the anomalous zone is used to determine the existence of evolutionary pressure in the transition zone. .

[0089] Extract the splitting state changes of the consensus region representation from the dynamic change trends of the representations of each region in the group cognitive topology. When multiple unconnected subclusters appear within the consensus region and the center distance between the subclusters exceeds a preset splitting distance threshold, it is determined that there is consensus region splitting evolution pressure.

[0090] Specifically, the system sets a preset splitting distance threshold. Extracting the split state changes representing the consensus region from dynamic trends, when the number of sub-clusters within the consensus region... Furthermore, when the minimum cluster center distance is greater than 0.3, it is determined that there is pressure for consensus region splitting and evolution. .

[0091] Extract the density change trend of the abnormal region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When the growth rate of the abnormal region sample density exceeds the second preset rate threshold, it is determined that there is pressure for the expansion and evolution of the abnormal region.

[0092] Specifically, the system sets a second preset rate threshold. Samples / day. Extract the density change trend of anomaly regions from dynamic trends; when the sample density growth rate of anomaly regions... At that time, it was determined that there was an abnormal region of expansionary evolution pressure. .

[0093] When at least two of the following conditions are met simultaneously: the transition region evolution pressure, the consensus region splitting evolution pressure, and the abnormal region expansion evolution pressure, the corresponding smallest cognitive unit is marked as the evolution pressure point.

[0094] Specifically, the system counts the number of values ​​of 1 among P1, P2, and P3. At that time, the smallest cognitive unit currently being analyzed is marked as an evolutionary pressure point.

[0095] Extract cognitive feedback data of the smallest cognitive unit associated with the evolutionary pressure point, analyze the historical friction response records and common misunderstanding patterns in the cognitive feedback data, and generate content evolution candidates. The content evolution candidates include: the smallest cognitive unit presentation order adjustment scheme, the supplementary explanation node insertion position, the interaction difficulty gradient reconstruction scheme, and the multimodal auxiliary resource replacement strategy.

[0096] Specifically, the system extracts cognitive feedback data from the smallest cognitive units associated with evolutionary stress points, analyzes historical friction response records, and statistically analyzes the effectiveness of each strategy. , The strategy targets specific misunderstanding patterns; analyzes common misunderstanding patterns, and statistically analyzes them. The top-3 patterns that appear most frequently. Content evolution candidates are generated according to the following rules: when... When established, a presentation order adjustment scheme is generated, moving the cell to the position after the adjacent cell with the lowest mobility; when When established, a supplementary explanation node insertion scheme is generated, inserting a connecting explanation node after the common preceding unit of the two split clusters; when Upon establishment, an interaction difficulty gradient reconstruction scheme is generated, breaking down the interaction steps of the unit into... Each sub-step simultaneously generates a multimodal auxiliary resource replacement strategy. resource replacement The most efficient alternative resource is among them.

[0097] Furthermore, following step S7, step S8 is also included: The content evolution candidates were verified by A / B testing in the test user group, and the occurrence rate of cognitive friction event identifiers and the change in path deviation characteristics of the test users were collected.

[0098] Specifically, the system deploys content evolution candidates to the comparative testing environment, and randomly divides test users into a control group and an experimental group, with a sample size of [missing information]. Friction occurrence rates were collected from two groups of users. Change in deviation from the overall path , and These are the five-dimensional path deviation feature vectors before and after content adjustment.

[0099] If the occurrence rate of the cognitive friction event identifier decreases and the change in the path deviation feature converges to the preset convergence interval, the verified content evolution candidate will be deployed as a new track science popularization content presentation scheme, and the cognitive feedback database and the preset path benchmark will be updated simultaneously.

[0100] Specifically, the system performs a one-sided independent samples test, and the test statistic... , and The figures represent the mean friction incidence rates for the experimental and control groups, respectively. and The variances of the two groups of samples, and For two groups of sample sizes. When And the experimental group The verification is deemed successful at that time. For significance probability, The standard deviation of the path deviation characteristics of historical high-performing users is taken. The system then deploys the validated content evolution candidates as new presentation schemes for track science popularization content, and simultaneously updates the cognitive feedback database, updating the expected understanding time by weighted average of new and old data. , The actual average dwell time of the experimental group in this unit; a new record is added to the common misunderstanding pattern library, including an automatically generated pattern number, a list of keywords, and a semantic embedding vector encoded by a pre-trained language model; the preset path baseline is also updated. The supplementary explanation nodes in the content evolution candidates are added as new nodes to the baseline path graph structure, and the new sequence scheme is marked as the current valid version.

[0101] According to a second embodiment of the present invention, the present invention claims protection for an iterative system for smart rail science popularization content based on immersive interaction, comprising: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, enable the processors to implement the immersive interactive smart rail science popularization content iteration method.

[0102] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0103] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for iterative development of smart rail science popularization content based on immersive interaction, characterized in that, Includes the following steps: S1, Time-align the multimodal user signals in the immersive interactive environment and extract basic cognitive state features, including attention representation, load representation and behavioral representation; S2, obtain the target understanding state pre-stored by the system, and perform cross-domain correlation between the attention representation, the load representation and the behavior representation and the target understanding state to generate a cognitive friction discrimination vector. The cognitive friction discrimination vector is used to characterize the degree of deviation between the user's cognitive state and the target understanding state. S3, compare and calibrate the cognitive friction discrimination vector with the preset friction threshold. If the deviation exceeds the preset friction threshold, trigger the cognitive friction event identifier. If it does not exceed the limit, it is marked as having normal cognitive function; S4, in response to the cognitive friction event identifier, query the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combine the cognitive friction event identifier and the cognitive feedback data to generate a friction response strategy and update the cognitive feedback data; S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate the original cognitive path, wherein the state outputs include the cognitive friction event identifier or the cognitive normal identifier. Obtain a preset path benchmark stored in the system, compare and calibrate the original cognitive path with the preset path benchmark, and extract path deviation features; import the original cognitive path and its path deviation features into the group cognitive dataset stored in the system, and store the path deviation features as metadata attributes associated with the corresponding samples. S6, perform statistical aggregation and regional weight modulation on the associated stored group cognitive dataset to generate cognitive state distribution features; The system obtains the pre-stored topology construction rules, compares and calibrates the cognitive state distribution features with the topology construction rules, and constructs a group cognitive topology. The group cognitive topology includes consensus region representation, transition region representation, abnormal region representation, and the dynamic change trend of each region representation. S7. Extract the depth change trend of the transition zone representation, the split state change of the consensus zone representation, and the density change trend of the abnormal zone representation from the group cognitive topology, and jointly determine the evolutionary pressure point; combine the evolutionary pressure point with the cognitive feedback data to generate content evolution candidates.

2. The method according to claim 1, characterized in that, In step S1, the multimodal user signals in the immersive interactive environment are time-aligned to extract basic features of cognitive state, including: The multimodal user signals include eye-tracking signals, physiological signals, and interactive operation signals. The multimodal user signals are uniformly mapped to a standardized time axis, and the signals from different sampling frequencies are time-aligned based on an interpolation algorithm to generate a synchronous multimodal signal stream. The eye-tracking signal in the synchronous multimodal signal stream is input into the attention parsing logic to extract the spatial attention distribution representation and the temporal attention persistence representation, and then combined to generate the attention representation. The physiological signals in the synchronous multimodal signal stream are input into the load analysis logic. Based on the skin conductance response variability index and the heart rate variability power ratio index, the cognitive load index and information processing rate are calculated and combined to generate the load characterization. The interactive operation signals in the synchronous multimodal signal stream are input into the behavior parsing logic to extract operation trajectory features and interface dwell features, and then combined to generate the behavior representation. The attention representation, the load representation, and the behavior representation are cascaded to form the basic features of the cognitive state.

3. The method according to claim 1, characterized in that, In step S2, the system pre-stores the target understanding state, and cross-domain correlates the attention representation, the load representation, and the behavioral representation with the target understanding state to generate a cognitive friction discriminant vector, including: The attention representation, the load representation, and the behavior representation are mapped to the same multidimensional feature space to construct a three-dimensional cognitive state tensor. The target understanding state vector is obtained from the system’s pre-stored target understanding state vector, which is generated by the expert-annotated understanding standard corresponding to the smallest cognitive unit or the feature center of historical high-understanding users. Calculate the spatial deviation between the three-dimensional cognitive state tensor and the target understanding state vector, and encode the spatial deviation and its directional components into the cognitive friction discrimination vector. The magnitude of the cognitive friction discrimination vector represents the degree of deviation, and the directional components represent the deviation dimension.

4. The method according to claim 1, characterized in that, In step S3, the cognitive friction discrimination vector is compared and calibrated with a preset friction threshold, including: The preset friction threshold is obtained, which includes a global baseline threshold and an individual adaptive threshold. The individual adaptive threshold is dynamically determined based on the statistical quantile of the distribution of the user's historical cognitive friction events or the upper limit of the confidence interval of the user's baseline cognitive state. Calculate the magnitude of the cognitive friction discrimination vector; If the modulus value exceeds the global baseline threshold and the individual adaptive threshold, the cognitive friction event identifier is triggered, and the friction type label is recorded. The friction type label includes attention deficit type, overload type, and behavioral deviation type. If the modulus value does not exceed both the global baseline threshold and the individual adaptive threshold simultaneously, it is marked as the cognitively normal identifier.

5. The method according to claim 1, characterized in that, In step S4, in response to the cognitive friction event identifier, the system queries the cognitive feedback data of the currently presented smallest cognitive unit pre-stored in the system, combines the cognitive friction event identifier with the cognitive feedback data to derive a friction response strategy and update the cognitive feedback data, including: The smallest cognitive unit currently presented is determined. The smallest cognitive unit is the smallest presentation granularity in the orbital science popularization content that has independent knowledge semantics and cannot be further divided. Each smallest cognitive unit has a unique unit identifier. The system queries the pre-stored cognitive feedback database based on the unique unit identifier to obtain the cognitive feedback data corresponding to the smallest cognitive unit. The cognitive feedback data includes expected understanding time, common misunderstanding pattern library, multimodal auxiliary resource index and historical friction response records. Based on the friction type label in the cognitive friction event identifier, the corresponding misunderstanding pattern in the common misunderstanding pattern library is matched, and the auxiliary explanation resources associated in the multimodal auxiliary resource index are called to generate the friction response strategy; The cognitive friction event identifier, the friction response strategy, and the response time are associated and recorded to generate a friction response record; The friction response records are merged into the historical friction response records of the smallest cognitive unit, and the statistical features in the cognitive feedback data are updated using a sliding time window.

6. The method according to claim 1, characterized in that, In step S5, the identifiers of the smallest cognitive units presented sequentially in a single experience and their corresponding state outputs are serialized and aggregated to generate the original cognitive path, including: Extract the unique unit identifier and its corresponding state output of each smallest cognitive unit presented sequentially in a single experience according to the time sequence, and construct a time-ordered state sequence. The state output is the cognitive friction event identifier or the cognitive normal identifier. The state sequence is encoded into a directed graph structure, where each node is a unique unit identifier of the smallest cognitive unit, the node attribute is the corresponding state output, and the edge is the transition relationship between adjacent smallest cognitive units, thereby generating the original cognitive path; Obtain the preset path benchmark stored in the system. The preset path benchmark is generated by the standard understanding path marked by domain experts or the cognitive path cluster center of historical high-performance user groups. The original cognitive path is aligned with the preset path benchmark by nodes and sequence comparison, and path deviation features are extracted. The path deviation features include node skip rate, node backtracking rate, path detour degree, friction node density, and deviation degree of the cognitive state at the end point. The original cognitive path and its path deviation features are imported into the pre-stored group cognitive dataset of the system, and the path deviation features are stored as metadata attributes of the sample and associated with the original cognitive path.

7. The method according to claim 1, characterized in that, In step S6, the stored group cognition dataset is statistically aggregated and region weighted to generate cognitive state distribution features and construct a group cognition topology, including: Kernel density estimation was performed on samples in the aforementioned group cognitive dataset according to cognitive state characteristics, and the distribution of friction frequency, load mean, and attention concentration in each local region was statistically analyzed. The samples are subjected to regional weight modulation, and the regional weights are multiplied and weighted according to the timeliness weight factor and the user experience level weight factor of the samples, wherein the timeliness weight factor decays as the interval between the sample collection time and the current time increases. The weighted samples are mapped to the cognitive state feature space, graph node connection relationships are constructed based on the cognitive state similarity between samples, and the region boundaries are divided based on the local density of the samples to generate the group cognitive topology. The consensus region is characterized as a set of samples in a high-density connected subgraph where the friction frequency is lower than the consensus threshold and the average load is in the middle range. The transition region is characterized as a set of boundary samples connecting different high-density connected subgraphs, with a local density lower than that of the consensus region and higher than that of the anomaly region. The abnormal region is characterized as a set of samples with a local density lower than the abnormal density threshold or a friction frequency higher than the abnormal friction threshold. The dynamic change trends of each region are obtained by differential calculation of the regional boundary displacement, sub-cluster number change and sample density change of the group cognitive topology under different time windows.

8. The method according to claim 1, characterized in that, In step S7, evolutionary pressure points are extracted from the group cognitive topology and combined with the cognitive feedback data to generate content evolution candidates, including: Extract the depth change trend of the transition region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When the rate of migration of the transition region sample to the abnormal region exceeds the first preset rate threshold, it is determined that there is transition region evolution pressure. Extract the splitting state changes of the consensus region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When multiple unconnected subclusters appear inside the consensus region and the center distance between the subclusters exceeds a preset splitting distance threshold, it is determined that there is consensus region splitting evolution pressure. Extract the density change trend of the abnormal region representation from the dynamic change trend of the representation of each region in the group cognitive topology. When the growth rate of the abnormal region sample density exceeds the second preset rate threshold, it is determined that there is pressure for the expansion and evolution of the abnormal region. When at least two of the following conditions are met simultaneously: the transition zone evolution pressure, the consensus zone splitting evolution pressure, and the abnormal zone expansion evolution pressure, the corresponding smallest cognitive unit is marked as the evolution pressure point. Extract cognitive feedback data of the smallest cognitive unit associated with the evolutionary pressure point, analyze the historical friction response records and common misunderstanding patterns in the cognitive feedback data, and generate content evolution candidates. The content evolution candidates include: the smallest cognitive unit presentation order adjustment scheme, the supplementary explanation node insertion position, the interaction difficulty gradient reconstruction scheme, and the multimodal auxiliary resource replacement strategy.

9. The method according to claim 1, characterized in that, Also includes: S8, A / B testing is performed on the content evolution candidates in the test user group to verify them, and the occurrence rate of cognitive friction event identifiers and the change in path deviation characteristics of the test users are collected. If the occurrence rate of the cognitive friction event identifier decreases and the change in the path deviation feature converges to the preset convergence interval, the verified content evolution candidate will be deployed as a new track science popularization content presentation scheme, and the cognitive feedback database and the preset path benchmark will be updated simultaneously.

10. A smart rail science popularization content iteration system based on immersive interaction, characterized in that, include: One or more processors; A memory having stored one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the iterative method for intelligent rail science popularization content based on immersive interaction as described in any one of claims 1 to 9.

Images