Painting guide method and device based on social influence field modeling and storage medium

Abstract

The invention discloses a painting guide method and device based on social influence field modeling and a storage medium. The method comprises the steps that visual focuses, public behaviors and environment topology of all audiences are sensed synchronously; constructing a dynamic social influence field based on behavior representation, and calculating the environmental pressure on the individual; by comparing independent cognitive baseline prediction with real-time observation interests, the social cognitive deviation degree is calculated in combination with environmental pressure, and an anti-consensus catalysis, cognitive anchoring enhancement or silent observation strategy is decided according to the social cognitive deviation degree; and finally, according to the strategy, tracing a social pressure source, and generating and pushing social repairable tour guide content with spatial narrative. According to the invention, quantitative management and active adjustment of the exhibition implicit social environment are realized for the first time, a guide system is converted into an active cognitive ecological balancer from passive information push, and the artistic thinking experience and cognitive autonomy of audiences are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of processing systems or methods for monitoring or prediction purposes, and specifically to a painting navigation method, device, and storage medium based on social influence field modeling. Background Technology

[0002] Existing personalized art tour systems treat viewers as isolated analytical units. Their core logic is to analyze individual viewers' trajectories, behaviors, and physiological data to infer their interests and push matching content. Even the most advanced solutions only consider the temporal evolution or delayed response of individual viewers' cognition, and their optimization goal is always to maximize the matching efficiency between individuals and exhibit information.

[0003] However, those skilled in the art generally overlook environmental factors: the "social influence" and "group dynamics" during the exhibition experience. An art exhibition is a typical socio-physical space; the viewer's experience does not occur in a vacuum. An individual viewer's attention, interests, emotional responses, and even the depth of their understanding of the artwork are unconsciously and significantly influenced by the behavior of other viewers (such as crowd gatherings, collective silence or exclamations, and the location and duration of others' attention). This influence can lead to two negative effects: first, the "conformity effect," where individuals may overestimate the value of an exhibit due to its popularity or ignore potentially valuable works due to lack of interest; second, "cognitive interference," where an individual's fragile, nascent independent insights may be drowned out or swayed by the mainstream reaction of the surrounding group. Existing technologies are completely powerless in addressing this because they lack the ability to perceive, quantify, and model the dynamic, non-contact, and implicit social influence networks within the exhibition hall.

[0004] At a deeper level, the spread of this social influence is not uniform; it is modulated by spatial distance, visual transparency, crowd density gradients, and individual "social reception sensitivity," forming a complex "social influence field." Currently, no technology has attempted to construct such a model in an exhibition environment, nor has any systematic thinking been undertaken on how to strategically counteract or utilize this social influence using guided tour technology to safeguard the independence and depth of individual aesthetic experiences. Summary of the Invention

[0005] The technical problem this invention aims to solve is: how to perceive and quantify in real time the dynamic, implicit social influence field generated by the spatial distribution and behavior of visitors within an art gallery, without relying on explicit social relationship data, and how to implement proactive cognitive guidance intervention for individual visitors based on this field model, thereby enhancing the overall cognitive diversity experience of the exhibition.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for guided art tours based on social influence field modeling, comprising the following steps: S100: Simultaneously perceive the individual visual focus trajectory of all visitors in the exhibition hall, the public behavior representation that can be perceived by others, and the topological characteristics of the exhibition hall environment; S200: Define individual influence sources based on the aforementioned public behavior representations, and construct a dynamic social influence field covering the exhibition hall by combining spatial attenuation and line-of-sight obstruction models; calculate the environmental social influence pressure value of each visitor's location; S300: Maintain an independent cognitive baseline model for each viewer. By comparing the interest score predicted by the model with the real-time observed interest score, and combining the environmental social influence pressure value, calculate the social cognitive bias, and make decisions based on the anti-conformity catalysis, cognitive anchoring reinforcement, or silent observation strategy. S400: Based on the selected strategy, generate and trigger socially restorative guided tour content that includes critical viewpoints, comparative analysis, or in-depth interpretation. The generation or triggering logic of the content is related to the spatial tracing of the main sources of social impact pressure.

[0007] Furthermore, in the above-mentioned painting guidance method based on social influence field modeling, in step S200, the dynamic social influence field strength at spatial point x is... The calculation formula is: in, For the audience u exist t The intensity of influence at any given moment is modulated by its real-time public behavior representation; for u The coordinates of the visual focus; For spatial points With focus The line-of-sight obstruction coefficient between them; σ is the influence space attenuation coefficient.

[0008] Furthermore, in the aforementioned painting guidance method based on social influence field modeling, in step S300, the social cognitive bias degree... The calculation formula is: in, To score the audience's observed interest at time t, The predicted interest score of the current exhibit i is determined by its independent cognitive baseline model. Let v be the environmental and social stress value experienced by v at time t, and ε be a smoothing constant.

[0009] Furthermore, in the aforementioned painting guidance method based on social influence field modeling, in step S300, when the social cognitive bias... Social and environmental stress levels When all values ​​exceed the corresponding threshold, an anti-conformity catalysis strategy is selected to generate questioning or comparative guided content designed to stimulate critical thinking; when... Low but When the level is extremely high, a cognitive anchoring reinforcement strategy is selected to generate in-depth interpretive content aimed at consolidating independent cognition.

[0010] Furthermore, in the above-mentioned painting guidance method based on social influence field modeling, in step S400, when generating social restorative guidance content, the spatial clusters corresponding to the main influence sources that constitute the current audience's high environmental social influence pressure and their historical exhibits are traced, and the historical exhibits are compared with the current exhibits at the content level to construct an implicit narrative that spans the exhibition space.

[0011] Furthermore, in the aforementioned method for guided art tours based on social influence field modeling, the intensity of micro-reactions in the public behavior representation... It is calculated by integrating the detection results of exclamatory sounds by lightweight real-time audio analysis and the results of posture change amplitude analysis based on computer vision.

[0012] Furthermore, in the aforementioned painting guidance method based on social influence field modeling, the independent cognitive baseline model... It is a personalized interest prediction model trained based on the behavioral data of audience v during periods of low environmental social influence stress in historical exhibition visits.

[0013] The present invention also protects an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the above-mentioned painting guidance method based on social influence field modeling.

[0014] The present invention also relates to a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-mentioned painting guidance method based on social influence field modeling.

[0015] The beneficial effects of this invention are as follows: For the first time, quantitative modeling and proactive management of the implicit social environment of exhibitions have been achieved: the S200 system integrates physical space topology and fine-grained audience focus behaviors through a field theory model, creatively transforming intangible social pressure into a computable continuous field. This forms a cross-disciplinary synergy with the "social influence theory" in social psychology, giving the system a "sixth sense" of perceiving group dynamics—an unprecedented breakthrough in this field.

[0016] This represents a paradigm shift from "serving individuals" to "regulating group-individual interactions": the independent cognitive baseline model and social cognitive bias calculation in step S300 constitute the core judgment mechanism. By comparing the "expected independent self" with the "actual self influenced by the environment," the system can accurately diagnose whether the cognitive state is "assimilated" or "adhered to." This works closely in synergy with the differentiated guidance strategy (anti-conformity / anchoring reinforcement) in S400, transforming the tour guide system from a passive information provider into a proactive "cognitive ecological balancer," dedicated to protecting individual cognitive diversity within the group atmosphere.

[0017] It creates a deep narrative experience based on spatial social dynamics: the spatial narrative generation in S400, based on tracing the sources of social influence, transforms the intangible tension of opinions among different groups of visitors in different locations within the exhibition hall into explicit dialogue content that guides visitors' reflection. Essentially, this uses technological means to construct the entire exhibition hall as a dynamic and immersive experimental field for social thought. Visitors receive not only knowledge but also insights into their own place in the social context, greatly enhancing the critical thinking and participation in the exhibition. This effect is a direct manifestation of the synergistic effect of spatial behavior perception, influence field tracking, and the generation of adversarial content. Attached Figure Description

[0018] Figure 1 The flowchart below illustrates a specific embodiment of the present invention involving a painting navigation method based on social influence field modeling. Detailed Implementation

[0019] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0020] Please refer to Figure 1 The present invention relates to a method for guiding visitors through paintings based on social influence field modeling, comprising the following steps: S100: The system synchronously collects the following data from all audience members present through a deployed sensor network (with audience consent obtained before entry and data collection): Individual focus trajectory: Based on head posture and gaze estimation, determine the three-dimensional coordinates of each visitor u's actual visual focus in the exhibition hall at time t. ; Public behavior representation: Extracting the behavioral feature vector of audience u at time t that can be perceived by neighboring others, including: Stationary state (moving / stationary); Microscopic response intensity: derived by combining lightweight audio analysis (such as interjection detection) with the amplitude of posture changes (such as leaning forward); Social focus direction: The pattern of their gaze deviating from the current focus and scanning other audience or crowd areas; Environmental topology features: Real-time calculation of the spatial visibility matrix of the exhibition hall, and real-time audience density in each area; In the above embodiments, S100 specifically includes: S101: Multiple sensor nodes integrating RGB-D cameras and thermal imaging modules are deployed in a hexagonal grid on the ceiling of the exhibition hall. Each node's coverage area overlaps with adjacent nodes by 30%, ensuring no blind spots. The RGB-D cameras provide color images and depth information (accuracy ±5mm), while the thermal imaging module helps distinguish individuals within dense crowds. The exhibits are concealed around each exhibit area; an acoustic array consisting of four miniature omnidirectional microphones is deployed with an array spacing of 25cm, a sampling rate of 48kHz, and a frequency response range of 50Hz-20kHz. Ambient light sensors, temperature and humidity sensors, and low-power millimeter-wave radar nodes were deployed in key locations in the exhibition hall. All sensors are synchronized with the master clock via a wired PTP (Precision Time Protocol) network with a time deviation of less than 1 millisecond. Each data packet carries a precise timestamp and sensor ID. S102: Individual focal trajectory acquisition utilizes an improved DeepSORT algorithm, fusing visual features from RGB images, body temperature features from thermal imaging, and micro-motion features from millimeter-wave radar to assign a unique and persistent ID to each viewer; identity feature update formula: Where λ = 0.95 is the forgetting factor; Let be the multimodal identity representation vector at time t; : The identity representation vector of the previous time step; It is a visual feature vector extracted from an RGB image, representing the appearance features of the viewer; It is an infrared radiation feature vector extracted from a thermal imaging image; It is a micro-motion feature vector extracted from millimeter-wave radar signals; Using a cascaded convolutional neural network, the first stage detects 68 facial keypoints from the RGB image, and the second stage predicts the gaze direction based on the keypoints and head pose (calculated by the PnP algorithm). The gaze direction is represented as a unit vector. ,in The pitch angle, The yaw angle is used; the 3D coordinates of key head points are calculated using the parallax of multiple RGB-D cameras to obtain the eye position. (The three-dimensional spatial coordinates of the viewer u at time t, where the center of both eyes (or the center of one pupil) is located). View vector Transformation from an eye-centric coordinate system to a world coordinate system: Where R is the rotation matrix from the eye coordinate system to the world coordinate system; The focal coordinates are obtained by solving for the intersection of the line of sight and the 3D model of the exhibit; the surface of the exhibit is pre-modeled as a set of triangular meshes. The intersection problem can be formalized as follows: Represents the collection of exhibit surfaces Any candidate three-dimensional point in the; Indicates Starting from, along A straight line in direction; Point to the straight line Euclidean distance; Expanded expression: This involves finding the point on the exhibit's surface that is closest to the line of sight; if the minimum distance is greater than a threshold δ (set to 0.1 meters), it is determined that the viewer is not focused on any exhibit. Set to empty; S103: Public behavior representation extraction and quantification: For each audience member u, extract behavioral characteristics that can be perceived by neighboring others; Calculate movement speed: in The location of the body's center of mass (obtained by weighted average of key skeletal points); The determination of a stationary state is a binary variable: if <0.15m / s, =1 (i.e., judged as a stationary state), if ≥0.15m / s, =0 (meaning it is judged to be in a non-staying state); Using a nearby microphone array, the sound source is located using the GCC-PHAT (Generalized Cross-Correlation-Phase Transform) algorithm. If the sound source is located near the audience... u Within a 0.3-meter radius of the mouth area, the following characteristics of the audio segment are analyzed: Variance of fundamental frequency profile F0: Dynamic range of the first dimension of MFCC (Mel frequency cepstral coefficients): Rate of change of the zero-crossing rate: ΔZCR Audio response intensity: Where sigmoid is the normalization function, and the weights are obtained through training on labeled data; Extracting posture response features from the skeletal keypoint sequence: Upper body forward lean angle: For a moment World coordinates of the midpoint of the audience's shoulder; For a moment World coordinates of the midpoint of the spectator's hip; It is a vertically downward unit vector (direction of gravity); It is the Euclidean norm; Output range: ; Rate of change of fore-angle: The intensity of postural response is characterized solely by the rate of change of the upper body forward lean angle, reflecting the sudden forward leaning movement of the audience due to emotional arousal or cognitive engagement. The formula is as follows: : Empirical calibration coefficient, used to map the rate of change of angle to the interval [0,1); Only retain the positive change (leaning forward), ignore the backward movement; : Hyperbolic tangent function, output range [0,1), suppresses extreme values ​​and ensures smoothness.

[0021] Fusion reaction strength: in =0.5 is the dwell time enhancement coefficient, which shows that the reaction is more easily detected when at rest; Define social attention metric vector : in The optimal line of sight from the eye to the current focal exhibit; Calculate the angle between the line of sight and the head direction of other visitors in the exhibition hall. If the angle is less than 15 degrees and lasts for more than 1 second, then... =1, otherwise =0; Calculate the similarity between the direction of the line of sight and the direction of the nearest crowd cluster center (obtained via the DBSCAN algorithm): :audience u In time t Attention level to the crowd (scalar, range [-1, 1]); : The coordinates (3D vector) of the center point of a population cluster; Clusters: The collection of all detected crowd clusters (obtained through clustering algorithms such as DBSCAN); Final social media following status: Take the category corresponding to the largest component.

[0022] S104: Dynamic modeling of environmental topology features; Basic visibility matrix Pre-calculation: The entire 3D space of the exhibition hall (the passable and observable space remaining after deducting the volume of walls and fixed obstacles) is divided into a uniform voxel (3D pixel) mesh; a resolution of 0.2 meters is chosen, which is a trade-off—too coarse will result in a loss of accuracy (e.g., inability to accurately represent narrow passages), while too fine will exponentially increase computational and storage costs; resulting in a set of N voxels. Each voxel represents a small cubic spatial unit; the precise 3D CAD model or point cloud model of the exhibition hall is loaded, in which all fixed obstacles (walls, columns, permanent partitions, large fixed booths) are modeled as closed triangular meshes. For each pair of voxels in the voxel set ,(in ), calculate its visibility : Get voxels and center point coordinates and Define a line segment connecting two points. ; Use ray casting algorithm to determine line segments Does it intersect with the triangular mesh of any static obstacle? If any intersection is found and that intersection is located between the two endpoints of the line segment (not the endpoints), it is determined to be invisible. If there is no intersection, then ; If there is an intersection, then ; Therefore, only the upper or lower triangular part can be calculated and stored to save space; Obstruction caused by moving obstacles (such as other visitors or temporary display boards) is calculated as follows: Where MO is the set of moving obstacles. Let k be the bounding box of obstacle k. =0.5 is the shading attenuation coefficient; The continuous density field is calculated using the kernel density estimation (KDE) method: in For the Epanechnikov kernel function, =1.2 meters is the bandwidth, =2 is the dimension. For the total number of viewers, Represents any three-dimensional spatial coordinate point within the exhibition hall space; For each exhibit i, define its effective area of ​​influence. (A circular region with a radius of 2.5 meters), the instantaneous density of this region is: In the above embodiments, multimodal identity binding ensures the continuity of individual trajectories in dense crowds, laying the foundation for building long-term behavioral models. The precise combination of gaze estimation and 3D focus calculation not only reveals where the audience is looking, but more importantly, what content-level information they are seeing, which is fundamental for judging cognitive states. Multi-sensor fusion of acoustic and posture responses overcomes the susceptibility to interference inherent in single-modal approaches (e.g., unreliable audio in noisy environments, limited vision in dark environments), improving the robustness of response intensity estimation. The dynamic visibility matrix breaks through the limitations of traditional static environment models, enabling social influence propagation models to reflect changes in gaze obstruction caused by crowd movement in real time, making it more consistent with real-world social scenarios. Continuous density field estimation more accurately reflects the shape and gradient of crowd aggregation than simple counting, providing high-quality input for influence field calculation.

[0023] Step S200: Construction of a dynamic social influence field and estimation of individual receptivity; Constructing a dynamic "influence field" based on physical space and behavior; Definition of influence source: Each viewer u is considered a dynamic "influence source"; its influence intensity Modulation by its own behavioral characteristics, especially A high state of residence will enhance its strength as a source; Influence Field Propagation Model: Defined at point x in space, the total social influence field strength from all viewers. A superposition field of spatial attenuation and line-of-sight modulation: Among them, the spatial decay of the Gaussian kernel simulation influence. The focal point of x and source u The line-of-sight occlusion coefficient; this formula means that the indirect influence of a viewer on someone far away or unable to see their actions is weak; Individual influence receptivity calculation: For audience v, the environmental and social influence pressure they experience at time t. The field strength at its location Sensitivity to social reception The product of: ; This is a learned personal trait parameter, initially set to 1, and updated online as follows: when the behavior of audience v (such as dwell time decision, reaction) is consistent with... When historical values ​​show a high statistical correlation, An upward adjustment in the value indicates that it is more susceptible to environmental influences; In the above embodiments, S200 specifically includes: S201: Definition of Multidimensional Influence Sources and Dynamic Energy Level Mapping; Each audience member is modeled as a "social energy source" with time-varying properties, whose state is described by the following tensor: Behavioral energy items: in, =0.7 is the dwell time enhancement coefficient (the influence is more easily received when stationary). =0.3 is the sensitivity coefficient for rate of change (sudden changes in response are more noticeable). and From S103; Spatial saliency term: in, =2.0 meters is the social distance constant, which reflects the visual psychology principle that "lonely individuals are more conspicuous than individuals in a crowd"; Topological centrality term: Calculates the betweenness centrality of the audience in the real-time line-of-sight network. in, The number of all valid sight paths between exhibits i and j. For the process The number of paths; this value is calculated using a real-time constructed audience-exhibit bipartite graph; Final source strength: Where σ is the sigmoid normalization function, and the weights α=0.5, β=0.2, χ=0.3 are optimized online through reinforcement learning; Define the social momentum vector of the source: in, and For the audience and its neighboring audience The location of the body's center of mass; The audience At any moment The social momentum vector (which is a direction vector that represents the direction of the influence of the local crowd flow trend on the audience). For the audience of Nearby audience members gathered. =5; Weight: =3 meters, For the audience and The angle between the lines of sight; This vector field encodes the flow trend of local crowds, giving the influence directionality, similar to the magnetization direction in an electromagnetic field. S202: Field propagation modeling in non-uniform media; We construct a social field equation inspired by Maxwell's equations, but introduce nonlinear corrections: Basic Poisson equation: in, Source density distribution; Time-varying medium parameters This parameter encodes the "resistance" of the environment to the spread of influence: =1.0 is the basic medium constant. =0.5 is the damping coefficient for population density (the denser the population, the more difficult it is for the influence to spread far). =0.3 is the line-of-sight damping coefficient. The average line-of-sight occlusion rate at position x (from S104); The Heaviside step function has the following standard definition: ; Boundary conditions: On the wall (Neumann condition). At the exit (Dirichlet condition), the simulated influence does not penetrate the wall and dissipates at the exit; The time progression uses the Crank-Nicolson implicit scheme to ensure numerical stability. Where A is the discrete Laplace operator matrix; The gradient of a scalar potential field determines the strength and direction of the influence flow: A curl correction is introduced to simulate the "vortex effect" of social impact (the circulation of crowds around popular exhibits): in An auxiliary potential field based on the eddy current of crowd flow; S203: Attention Focus Weighting and Semantic Enhancement; Elevating the spread of influence from the purely physical space to the semantic space: Each exhibit i has its semantic vector. (d=300, extracted using BERT based on work description, artist style, and era characteristics); Define the semantic attention weight of audience u to exhibit i: in The real-time interest vector of the audience (obtained by exponentially decaying average of the semantic vectors of their historical focal exhibits). =0.1 represents the temperature parameter; Semantic correction of source terms: in The total attention received by exhibit i within the time window [t-ΔT, t] is the total number of times all visitors focus on it (the integral of the total number of times all visitors focus on it). This gives audiences who are highly interested in popular exhibits a stronger "theme-specific influence".

[0024] S204: Dynamic Calculation and Learning of Individual Acceptance; Receiver pressure calculation; Basic pressure: in Let v be the local density (within a radius of 2 meters). This is the maximum permissible density for the exhibition hall; Line-of-sight alignment modulation introduces a receiver efficiency factor: in, Let v be the unit vector of the line of sight. for The horizontal angle between v and the direction from which the main source of influence is located. =π / 4; Ultimate environmental stress: ζ=0.4, more susceptible to environmental influences when stationary; Online learning on social reception sensitivity: Define behavioral-stress consistency metrics: Calculate the observation interest score Environmental pressure in the past =Pearson correlation coefficient within a 300-second window; Sensitivity parameter update law: =0.01 is the learning rate, and tanh ensures smooth updates; when Updates are paused when the value is less than 0.3 to avoid noise-induced learning. S205: Interface for field visualization and real-time monitoring; The system provides curators with a real-time dashboard, displaying: Social potential energy topology diagram: displayed in heat map form Spatial distribution, superposition Streamlines; Key source identification: tagging The top 10% of viewers are identified, and their social momentum vectors are displayed. Stress distribution statistics: Real-time display of audience stress levels Histogram, marking high-pressure groups ( Location clustering (>threshold); Semantic field analysis: Principal component analysis is used to project the high-dimensional semantic field onto 2D to show the distribution of the influence of different themes in the exhibition hall.

[0025] Step S300: Independent cognitive tendency assessment and guidance strategy generation; This step compares external social pressures with internal cognitive states to determine whether intervention is necessary. Independent cognitive baseline modeling: Maintain a baseline for each viewer v based on their personal historical behavioral data (excluding high-risk individuals). Light interest prediction model trained during a specific time period The model predicts the "expected independent interest score" of viewer v for the current exhibit i in the absence of social interference. ; Cognitive Bias Detection: Calculating the Degree of Social Cognitive Bias : in, The "observation interest score" is derived by inferring the real-time behavior (duration of stay, micro-reactions) of the audience v in front of the current exhibit i. A high value indicates that the audience’s current interest is significantly inconsistent with their personal baseline prediction, and this inconsistency occurs when they are under high social influence pressure, which is likely due to conformity or interference. Guiding strategic decision-making: Based on and The system dynamically selects from the following strategies: Strategy A (Anti-conformity catalyst): When High and High (significant deviation from baseline under strong pressure) suggests that viewers may be abandoning independent judgment; trigger strategy: push questioning or comparative content. For example, for a popular exhibit, the push content could be: "This work has received much attention, but critic A has pointed out that its composition is too directly related to the traditional paradigm Y. Do you think this is a tribute or a lack of innovation?" Strategy B (Cognitive Anchoring Reinforcement): When Low but Extremely high (maintaining baseline even under intense pressure), suggesting the viewer may be engaging in difficult independent cognitive processes, requiring support; Trigger strategy: Push in-depth analysis or background-reinforcing content to provide "ammunition" for their independent viewpoint. For example: "You lingered over this obscure work for a long time. Indeed, its use of color was ahead of its time, subtly connecting to the Z-style art that would become popular ten years later..." Strategy C (Silent Observation): When or When the temperature is low, no special intervention will be implemented; push notifications will be sent according to the usual personalized logic. In the above embodiments, S300 specifically includes: S301: Establishment and dynamic maintenance of independent cognitive baselines; Define a "pure observation window": filter historical behavioral data of audience v that meet the following criteria. and Time segments; Take the 30th percentile of the overall pressure value distribution. 3 people per square meter; For each exhibit contact event within a clean window, extract the feature vector: Where duration is the length of stay (log-standardized). is the rate of change of reaction intensity, focus_stability is the proportion of time the gaze lingers in the core area of ​​the exhibit, and exploration_ratio is the coverage of different parts of the exhibit. The model is trained using the Gradient Boosting Decision Tree (GBDT) framework. Label The overall interest score is calculated using time decay weighting: in =0.98 is the attenuation factor. =0.6, Calculated from eye-tracking data (fixation duration and saccade amplitude); The model output is a predicted interest score for any feature vector f. ; The model is updated incrementally every 24 hours, and a sliding window is used to retain the clean data of the most recent 7 days. When a visitor v comes into contact with a new exhibit i, the preview features from the first 3 seconds are extracted in real time. (Including approach speed, initial gaze pattern, etc.); Calculate the expected independent interest score: Where κ=0.2, Semantic similarity calculation: This is the long-term interest vector of audience v (an exponentially weighted average of the semantic vectors of historical focal exhibits). For current exhibits semantic vector; S302: Real-time solution for multi-dimensional observation of interest; Constructing real-time observation feature vectors : Standardized stay duration The median dwell time of exhibit i in history. Duration of stay; Recent reaction intensity weighted sum, For triangular window functions, , refer to S103; : Gaze direction variance normalization The angular variance of the line of sight direction over the past 5 seconds; Micro-expression activity ratio, detected by the amplitude of facial key point movements; The amount of micro-expression activity represented by the movement of facial key points within a specified time window; The total amount of facial key point movement (including micro-expressions and macro-expressions) within the same time window. Using pre-trained neural networks Mapping (the fully connected layer structure [4,8,4,1]): Where σ is the sigmoid function, the network is trained on a large labeled dataset, and the labels are derived by experts based on comprehensive behavioral evaluation; Introducing cognitive consistency verification, we calculate the internal consistency index of observed features: in and for and The normalized version; when consistency < 0.5, multi-time-window smoothing is triggered: ; S303: In-depth calculation and attribution analysis of social cognitive bias; Basic deviation calculation: Pressure-deviation coupling correction: in The stress normalization factor is the 75th percentile of historical stress values. Time decay factor: =300 seconds, The time for first contact with exhibit i; This formula demonstrates that social pressure amplifies the deviation between observed behavior and the baseline; and that the effect of this deviation diminishes over time.

[0026] Calculate the directional component of the deviation: Positive values ​​indicate "overreaction" (actual interest is higher than expected), while negative values ​​indicate "underreaction". Social field source analysis: For each high-influence source Calculate its contribution weight, where For spatial distance, for The main focus is on the semantic vectors of the exhibits; Identify the dominant influence type: like This is labeled as "group pressure-driven". If there is a single If the weight is >0.4, it is marked as "opinion leader influence type"; like Marked as "Theme Resonance Type"; S304: Three-State Decision-Making Logic and Strategy Selection; Establish a three-dimensional decision coordinate system: X-axis: Social cognitive bias (Normalized to [0,1]); Y-axis: Environmental pressure (Normalized to [0,1]); Z-axis: Deviation direction The absolute value; The decision space is divided into three regions: Strategy A (Anti-conformity catalyst zone): Confidence level = ; Inferring that viewers may be abandoning independent judgment; trigger strategy: push questioning or comparative content. For example, for a popular exhibit, the push content could be: "This work has received much attention, but critic A has pointed out that its composition is too directly related to the traditional paradigm Y. Do you think this is a tribute or a lack of innovation?" Strategy B (Cognitive Anchoring Reinforcement Zone): Confidence level = ; It can be inferred that the viewer may be engaging in difficult independent thinking, which requires support; triggering strategy: push in-depth analysis or background-enhancing content to provide "ammunition" for their independent viewpoint; for example: "You lingered for a long time at this obscure work. Indeed, its use of color was ahead of its time, and has a subtle connection to the Z-style movement that will become popular ten years later..." Strategy C (Silent Observation Zone): Other cases, confidence level = ; Without special intervention, push notifications will be sent according to the usual personalized logic. in The threshold value was determined experimentally. Strategy C is the default rule, and its confidence score is defined as the complement of the parts not covered by the first two rules. It is usually calculated as follows: The activation strength (confidence) of each rule is obtained by taking the minimum membership degree of each variable in its preconditions; Compare the confidence scores of all rules and select the policy corresponding to the rule with the highest confidence score as the output: To ensure the robustness of the decision-making process, a confidence threshold is set. : Anti-conformity strength factor: Anchoring depth factor: These factors are then passed to the S400 to control the intensity and depth of the guided tour content; S305: Real-time diagnostic dashboard and feedback loop; Provide curators with a real-time view showing each audience member's: location in a three-dimensional decision space (dynamic point); time-series comparison curve of baseline prediction and observed scores; source map of social stress; current strategic decisions and confidence levels; Collect behavioral feedback after strategy implementation: calculate the time period after intervention. Changes in interest scores within ; Evaluate the effectiveness of the strategy: ; It is a small smoothing constant (e.g., 0.01); Fine-tuning decision thresholds through online learning: if ,but if ,but in =0.1 is the effect threshold. =0.05 is the adjustment step size; Step S400: Generation and delivery of socially restorative content based on spatial narrative; To implement the guidance strategy, non-standard content needs to be generated; Content library expansion: In addition to regular explanations, "socially restorative" content tuples are pre-set or generated in real time for each exhibit, including: critical viewpoints, minority interpretations, in-depth technical analysis, and historical controversies; Spatial narrative triggering: The timing and content of push notifications depend not only on the current state of exhibit i and visitor v, but also on the sources of social impact; the system can trace current high-level... The main contributing sources (i.e., which other audiences or groups of people) are identified, and it is determined whether these sources have gathered at exhibit j, which is thematically opposed to or competes with exhibit i. If so, the generated content can proactively mention j, constructing an implicit spatial debate narrative. For example: "The realistic portrait (i) you are admiring right now forms an interesting contrast with the highly sought-after abstract group portrait (j) on the other side of the exhibition hall; the latter seeks emotional catharsis, while the former returns to classical tranquility and eternity; which expression do you prefer?" Push notification execution: Guided push notifications use a gentler triggering mechanism (such as requiring the audience to make slight active exploration behaviors, such as clicking the screen or staring at a certain detail) to reduce the feeling of imposition; In the above embodiments, S400 specifically includes: S401: Construction and dynamic indexing of multi-granularity content knowledge base; Each content fragment is defined as a seven-tuple: = type∈{basic description, technique analysis, historical background, critical perspective, comparative analysis, personal reflection prompts, related retrospective narrative, questioning guidance, in-depth reinforcement, spatial narrative}; depth∈[1,10]: Cognitive depth level, assessed by an expert annotation team based on the amount of prior knowledge required for comprehension; stance∈[-1, 1]: Degree of stance polarization, -1 for complete criticism, 1 for complete approval, 0 for neutrality. target_vector∈ Target audience feature vector, dimensions include [professional knowledge level, cognitive style (analysis / intuition), emotional sensitivity, time budget, social orientation]; semantic_embedding∈ Content semantic vectors generated based on the ALBERT model template_structure: Defines the syntactic template and variable slots for the content; Create content fragments and Multiple relationships between them: ={prerequisite_of,contradicts,complements,contextualizes,simplifies}; and It refers to any two distinct content fragments within the knowledge base; Relationship strength is calculated based on: semantic vector cosine similarity and the co-occurrence frequency of audience interaction data (when audiences make sequential requests). and Logical dependencies (time-based) and expert-annotated; S402: Content synthesis engine based on cognitive state; 1. Anti-conformity catalyzes the generation of content (corresponding to strategy A); Input: The current state of the audience v The current exhibit i primarily influences source information; Content selection algorithm: a. Retrieve candidate sets: b. Calculate the fit score: in: The greater the deviation, the more in-depth the content should be. The intensity factor determines the degree of criticism; Measurement of contradiction reinforcement: This contrasts with the theme of the source of influence; The semantic vector of the guided content segment (such as a critical viewpoint) encodes the theme, stance, and argumentation logic of the content; For the source of influence u The semantic vector of the exhibits that an opinion leader or a group focuses on represents the mainstream social influence or popular viewpoint currently exerted on the target audience. This value measures the directional similarity between two high-dimensional semantic vectors, typically ranging from [0, 1] (assuming the vectors are normalized and located in the positive quadrant). The closer the value is to 1, the more semantically similar the content c is to the exhibits (representing mainstream viewpoints or themes) that the source u focuses on; c. Adjust weights based on attribute_type: If it is "group pressure-driven": increase Weight (emphasizing opposition); If it is an "opinion leader influence type": increase Weighting (for specific viewpoints); Variable instantiation: Extracting clustered exhibits from the social field to identify key influencing sources. j ; Fill in the template variables: [Opposing Exhibits] = j.name, [Mainstream Viewpoints] = description of conformity behavior, [Alternative Perspectives] = c.content; Generate a guiding question: "When most people focus on characteristic X of [the opposing exhibit], have you noticed characteristic Y of this work?" 2. Cognitive anchoring reinforces the generation of content (corresponding to strategy B): Input: Independent interest baseline of audience v Real-time awareness of status Current exhibits i Content selection algorithm: a. Retrieve candidate sets: b. Calculate cognitive continuity score: in: Search audience historical interest vectors The moment most relevant to c in the middle. =0.95 is the attenuation factor. =0.6; Choose content that matches your expected interests in depth; c. Cognitive scaffolding construction: If difficulty in understanding the current content is detected in the audience (through micro-expression confusion indicators), a simplified pre-content is automatically inserted. ,satisfy `prerequisite_of` represents a logical dependency between content segments. It indicates a relationship between two navigation content segments. There is a "precondition" relationship between it and c; Personalized argument chain generation: Based on the viewing history of audience v, a personalized cognitive development narrative is constructed: "You previously showed interest in [historical exhibit k], and its [feature A] develops into [feature A'] in this work..."; Using a Bayesian inference model, an explanatory chain is generated: P(understanding the current work | having understood historical work k) > θ; 3. Generation of spatial narrative content (a special form of strategy A / B): Input: Spatial distribution of influence sources obtained from social field tracing and its featured exhibits ; Current exhibits: ; Narrative generation algorithm: a. Constructing a spatial-semantic conflict map: =0.7: Contrast threshold; Semantic contrast calculation: b. Generate a dialectical narrative framework: Exhibits on the left side of the exhibition hall j The group represents an [artistic proposition A], and the [exhibit] you are viewing i This presents [opposing claim B]... c. Insert metacognitive prompt: "This spatially contrasting layout may be a deliberate arrangement by the curator, inviting viewers to establish their own stance between the two." S403: Dynamic Triggering Mechanism Based on Cognitive Rhythms 1. Cognitive state window analysis: Calculate the frequency domain characteristics of real-time cognitive load: Capture audience v over a period of time (For example Real-time cognitive load value sequence within 30 seconds This load value is a scalar calculated in real time in S300, which integrates multiple dimensions such as micro-response intensity, gaze stability, and social pressure.

[0027] Extract dominant frequency This reflects the rhythm of cognitive processing (e.g., slow thinking vs. rapid scanning). High frequency dominance (e.g.) >0.5 Hz): This may indicate that cognitive processing is in a state of rapid scanning, evaluation, or decision-making. The viewer may be quickly comparing details and making category judgments, exhibiting a fast cognitive pace.

[0028] Low frequency dominant (e.g.) <0.2 Hz): This may indicate that cognitive processing is in a state of deep integration, contemplation, or insight preparation. Viewers may construct complex mental models during prolonged staring, with a slow and profound cognitive pace. Define cognitive phase: By integrating the dominant frequency over time, a continuous cognitive phase is obtained. The phase value cycles between 0 and 2π, describing the specific position of the cognitive rhythm within the cycle. This is the moment when visitors enter the area affected by the current exhibit; The timing of the push must meet the phase alignment condition: ; phase Discretize the continuous cognitive oscillation cycle into angles ranging from 0 to 2π. Studies (such as the "phase-locked" hypothesis) show that the brain's sensitivity to external stimuli and its processing efficiency vary with the phase of its internal oscillations. Highly sensitive phase (e.g.) ≈ 0): The cerebral cortex is in an excited state, which makes it highly efficient at receiving external information and easy to encode it; Low-sensitivity phase (e.g.) (≈3π / 2): The brain is in an inhibitory or internal processing phase, during which external interference may disrupt the ongoing internal cognitive process; 2. Multiple conditions combined to trigger judgment: Construct the trigger decision function: in: It is the Sigmoid function, which maps the result of the linear combination to the (0, 1) interval as the final trigger probability; These are the weight coefficients of each item, usually obtained through offline learning or reinforcement learning optimization, satisfying... ; It is a trigger threshold bias, used to control the overall triggering tendency of the system (the higher the threshold, the more conservative the triggering). Readiness level: This refers to the cumulative time spent in front of the current exhibit i; The minimum cognitive preparation time required for exhibit i is preset based on the exhibit's complexity and content type. The ratio ensures that visitors have a minimum observation time. This is an indicator of gaze stability (such as the variance of gaze direction over the past 3 seconds). This ensures that the viewer's attention is focused on the exhibits, rather than wandering off. Only when the audience stops and observes attentively for a period of time are they considered "ready" to receive deeper information; Acceptability: Real-time cognitive load estimates from the S300; The ideal cognitive load level (an empirical value, such as 0.4) is when information processing efficiency is highest. The valid range of the load value (e.g., 0.2) is used for normalization; This function is a... The peak value is an inverted V-shaped function. Receptivity is highest when the real-time load is close to the ideal value; receptivity decreases when the load is too low (possibly due to boredom and inattentiveness) or too high (possibly due to cognitive overload); this ensures that push notifications occur at the "sweet spot" when cognitive resources are sufficient and not over-utilized. Context suitability: The local audience density (people / square meter) is considered; high density can easily lead to distraction. The ambient noise level (decibels) is derived from the microphone array. The preset maximum tolerance density and noise level; This simulation examined the negative impacts of social interaction and noise interference on information reception. When the surroundings are noisy and crowded, suitability decreases because content delivery is more likely to be overwhelmed or interrupted. 3. Progressive triggering strategy: Level 1 Trigger (Mild): When Trigger(t) > 0.6, display a non-intrusive cue icon on the AR interface; for example, display a gentle, non-textual visual cue icon at the edge of the viewer's augmented reality (AR) glasses' field of vision, or in a non-core area of ​​the handheld device's screen. Examples include a slightly pulsating question mark "?", an abstract thought bubble graphic, or a decorative symbol consistent with the current art style. Level 2 trigger (moderate): If the viewer stares at the icon for more than 2 seconds, play a short audio prompt; for example, play a very short (about 3-5 seconds) guiding audio clip; the content does not contain specific information, but only guiding words, such as: "There is a different perspective on this work...", "Some background may add interest...", or a soft prompt followed by "Please listen"; Level 3 trigger (depth): If the viewer makes a clear interactive gesture (such as raising their hand or nodding), push the complete content; for example, a 45-second "spatial narrative" audio or AR overlaid text; Abandon triggering: If <-0.1 (receptivity drops rapidly), cancel all triggers; S404: Multimodal content presentation and cognitive load optimization; 1. Presentation channel adaptation algorithm: Based on real-time cognitive load allocation information channels: The CognitiveCost model is constructed based on experimental data from the cognitive dual-task paradigm. Real-time load value From S300, and all currently available content presentation channels m; invoke the predefined cognitive cost model. ; Calculate the additional cognitive burden that using this channel to present content is expected to bring to the audience under this cognitive load level, and select the channel with the lowest cognitive cost as the presentation method for this content push; 2. Rhythm and pause control: Adjust speaking speed based on real-time audience comprehension feedback: base_rate is the base speaking speed, such as 180 words / minute, preset according to content type and exhibition style; comprehension_index(t) is a real-time comprehension index, a scalar that dynamically changes in the range of [-1, 1], reflecting the audience's level of understanding of the current content and the degree of pacing match. For example, the gain coefficient. = 0.3), the sensitivity of controlling speech rate to the comprehension index; When comprehension_index(t)>0: it indicates that the audience understands well and is actively following along. The system will appropriately increase the speaking speed (up to 1.3 * base_rate) to match their efficient information processing ability and avoid distraction caused by a slow speaking speed. When comprehension_index(t) < 0, it indicates that the audience may have difficulty understanding or their attention may be delayed. The system immediately slows down the speech rate (to a minimum of 0.7 * base_rate) to allow them more time for cognitive processing. When comprehension_index(t) ≈ 0: maintain the baseline speaking rate; Insert deliberate pauses after key concepts, duration: Use a basic pause duration, such as 0.8 seconds, to ensure a basic breathing interval between statements; concept_complexity is a complexity score for the next key concept to be announced, ranging from [0, 1]; this score is based on predefined knowledge graphs (e.g., "Perspective" = 0.3, "Deconstruction" = 0.9). This is the complexity-pause duration conversion factor, for example, 1.2 seconds; S405: Real-time effect evaluation and dynamic adjustment; 1. Multidimensional evaluation of the effects of short-term interventions: Define effect vectors: Specifically, this vector analyzes the short time window following a guided tour intervention (such as pushing a piece of questioning content) from five independent yet related dimensions. A quantitative snapshot of the overall effect within a 60-second timeframe (e.g., quantifies the effect). Environmental stress changes , The environmental social impact stress value calculated from S200; this value is used to measure the moderating effect of the intervention on the socio-psychological environment in which the audience is situated; reduced stress means that the intervention may have helped the audience psychologically "shield" or "transcend" the influence of the group; Successful interventions are often accompanied by This indicates that the system has constructed a temporary "cognitive shield" for the audience; Changes in social cognitive biases The social cognitive bias metric calculated from S303; this value measures the effectiveness of the intervention in addressing the core issue—whether the deviation between the audience's behavior and their true self (independent baseline) is reduced or increased. This is the core cognitive indicator for assessment; Successful intervention must meet the following requirements: This means a reduction in deviation; this is the gold standard for defining a successful intervention. Consistency gain: ; For observational interest, The independent interest baseline predicted by S301; It is a sign-sensitive measure of deviation change, which is more than It is more refined, directly quantifying whether the absolute distance between the observation interest and the individual baseline is narrowing or widening; The absolute distance was reduced, and the intervention helped the audience "be themselves again," with positive results. As the absolute distance increases, intervention may make the animal more "disoriented," resulting in negative effects. Cognitive novelty response: Measured using an optional lightweight EEG (electroencephalography) headset, measuring the relative change in the power of theta waves (4-8 Hz) in the prefrontal cortex of the viewer; novelty_response is the ratio of theta wave power after intervention to baseline power; for high-quality anti-conformity or anchoring content, a novelty_response > 1 (theta wave enhancement) is expected, indicating that the viewer is engaging in deep thinking, establishing new connections, or reconstructing existing concepts. 2. Optimization of reinforcement learning strategies: Each intervention is modeled as a Markov decision process. state action Indicates the next state after the action is performed; strategy_type: Decision strategy type, discrete action, selected from {anti-conformity catalysis, cognitive anchoring reinforcement, silent observation}; content_depth: Content depth level, continuous actions, range such as [1, 10], determines the cognitive complexity of the pushed content; trigger_timing: Trigger timing parameter, continuous action, which can be mapped to the bias term of the trigger decision function in S403. The adjustment amount; award Online policy network update using the Actor-Critic algorithm Critic Network - Value Function V(s): Input: state s; Output: A scalar that evaluates the expected total long-term reward that can be obtained by following the current policy in the current state s; Function: As a "mentor", it provides a more stable baseline for the updates of the Actor network, reduces learning variance, and accelerates convergence; Update: The value of the learned state is updated through temporal difference (TD) error; Actor Network - Policy Function π(a|s): Input: state s; Output: The probability distribution of action a (for discrete actions) or distribution parameters (such as the mean and variance of a Gaussian distribution for continuous actions). Function: As a "decision-maker", it directly outputs the action strategy to be taken based on the current state; Update: The "advantage function" A(s,a) = r + γV(s') - V(s) calculated using the Critic network is used for updating; A(s,a)>0: This indicates that action a performs better than average in state s, and the probability of this action should be increased. A(s,a)<0: This indicates poor performance and its probability should be reduced. The update direction is along The gradient direction of _θ log π(a|s) * A(s,a), where θ is the parameter of the Actor network. 3. The evolution mechanism of the content library: Calculate the effectiveness of content segments based on audience interaction data: It is an interactive positive feedback discriminant function that determines whether a single interaction between the audience u and the content segment c produces a positive effect; The input is the complete interaction log of a single push of content c to audience u, including the state s before the push, the multi-dimensional effect vector Effect after the push, and the subsequent behavior sequence. The decision-making logic is based on a predefined set of positive feedback rules. The rule set is a combination of multiple conditions, for example: Strategy A: Strategy B: Strategy C: The output is a boolean value; it returns True if any positive feedback rule is satisfied, otherwise False. The overall effectiveness score of the content segment is obtained by summing the weighted positive feedback from all viewers u who have interacted with content c. ; The system periodically (e.g., after closing time each day) calculates all content fragments in the content library. Scoring will be based on performance. Content that has been online for more than a certain period (e.g., 2 weeks) but still has fewer interactions than the minimum threshold (e.g., 5 times) will be considered "untested" and temporarily archived. Content with sufficient interaction data will be scored accordingly. Score sorting. Content snippets that consistently rank in the bottom 10% (and whose absolute score is below a certain threshold, such as 0.5) are marked as inefficient content, removed from the active push library, and moved to the "historical archive". When the system detects a lack of efficient content in the current active content library for a certain type of exhibit (semantic cluster) or a certain cognitive state (such as "slight positive deviation under extremely high pressure"), When the mean is low, fill in the blanks; Based on generative models (such as GPT-4): Input: Provide "seed" information, including: semantic description of the target exhibit, and a profile of the target audience (e.g., The scope, expected content type and stance (such as "critical viewpoint", "in-depth analysis"), and several currently most effective similar content as excellent examples; Generation: Based on the above conditions, the model generates a new batch of candidate content text; Template-based and rule-based: For content requiring high accuracy (such as historical facts) or a specific structure (such as comparative analysis), expert-predefined templates can be used, which are then automatically filled in by the system with entities and relationships retrieved from the knowledge graph.

[0029] In step S400, the output of S300 is Directly control the content selection function of S402 and Weight allocation in the S200; the social field constructed by S200 is not only used to calculate stress values, but also for its source tracing results. Directly used as input for spatial narrative generation in S402, it imbues the content with a realistic social context; the cognitive phase analysis in S403 is based on S300. Calculation ensures that the timing of push notifications matches the audience's internal cognitive rhythm, which is a key innovation to avoid cognitive interference; the selection of presentation channels is based on real-time cognitive load, and the load value itself is affected by the content being intervened and the presentation method. The system finds the best balance point through iterative optimization; the effect evaluation data of S405 not only optimizes the strategy selection, but also adjusts the content selection parameters through reinforcement learning, and even drives the evolution of the content library, forming a continuously improving ecosystem.

[0030] Specific examples: Exhibition Background: An art museum is hosting an exhibition themed "The Two Sides of Modernity: Abstraction and Realism." Major works include: Painting A (Abstract): "Order of Chaos", a large-scale abstract expressionist painting with intense colors and unrestrained brushstrokes, is the marketing focus of the exhibition and attracts a large number of visitors.

[0031] Painting B (Realism): "Contemplation by the Window", a hyperrealist portrait with extremely fine details and subtle emotions, located in a relatively quiet corner of the exhibition hall.

[0032] Painting C (Abstract): "Color Variations", a medium-sized color gamut abstract painting, is a complementary work to painting A.

[0033] Audience member: Mr. Lin, who has an architecture background, has a consistent interest in content related to structure, proportion, and fine craftsmanship based on his personal history data.

[0034] Step S100: Data Acquisition and Feature Extraction (25 minutes after the start of the exhibition); Sensor network capture: A ceiling-mounted RGB-D camera array detected Mr. Lin (ID: L001) 2.1 meters in front of painting A. A synchronized microphone array captured dense exclamations and shutter sounds in the area.

[0035] Focus trajectory reconstruction: Facial landmark detection: Mr. Lin's head posture is slightly tilted upward (15 degrees).

[0036] Line of sight estimation: its line of sight unit vector The focus is primarily on the central splashed area of ​​painting A.

[0037] 3D intersection calculation: The intersection point F_L(t) = (x: 10.2, y: 3.5, z:1.8) between the line of sight and the digital model of painting A is confirmed as the core area of ​​painting A.

[0038] Public behavior characteristics: Dwelling state: speed It is determined to be stationary. ).

[0039] Microscopic reaction intensity: Audio analysis: The fundamental frequency variance of Mr. Lin's interjections was determined from nearby microphone captures. (Lower).

[0040] Posture analysis: Upper body forward lean angle The degree of change is gradual; the trajectory complexity of key hand points is low.

[0041] Fusion computing: =0.2, =0.15, = 0.3.

[0042] Social attention: = 0.92 (high), and All are below 0.3, indicating a focus on exhibits.

[0043] Environmental topology: Real-time audience density in area A of the painting = 2.1 people / m 2 Area B of the painting = 0.4 people / ㎡. Mr. Lin has an unobstructed view of painting A. ≈0).

[0044] Step S200: Social Influence Field Modeling and Stress Calculation; Identifying the source of influence: Five viewers (e.g., U01-U05) in front of painting A exhibited a high level of reaction. >0.7 (taking photos and talking), is calculated as a strong influence source. Assuming the strength of viewer U01 is... =0.85, its focal point Also in painting A.

[0045] Calculate the electric field strength: Mr. Lin's location .

[0046] Calculate the influence contribution of U01 based on the formula: distance. = 1.5m, unobstructed, Gaussian kernel decay term ≈0.68.

[0047] The effects of all sources are superimposed, and high density is taken into account. The resulting propagation damping (increased μ) ultimately led to the calculation of the social influence field strength at Mr. Lin's location. = 0.76 (high).

[0048] Calculate individual reception stress: Mr. Lin's gaze direction The receiving efficiency is basically consistent with the main direction of the field strength. ≈ 0.9.

[0049] Its historical data shows social reception sensitivity = 1.1 (slightly higher than average).

[0050] Environmental and social influences stress: = 0.76 * 0.9 * 1.1 * (1 + 0.4 * 1) ≈ 1.05 (Standardized value, much higher than the threshold) =0.6).

[0051] Step S300: Independent cognitive tendency assessment and strategic decision-making; Invoking an independent cognitive baseline: The system retrieved Mr. Lin's pure observation history and his baseline model. The results show that for abstract works with wild brushstrokes and loose structure, the predictive interest is usually low (E≈0.3-0.4), while for realistic works with fine details, the predictive interest is high (E≈0.7-0.8).

[0052] For the current artwork A (abstract), the model is based on preview features. Based on semantic similarity, output the expected independent interest score. = 0.35.

[0053] Calculate observational interest: Real-time characteristics: dwell time ratio =1.2 (longer), reaction strength =0.3 (low), gaze stability =0.85 (high), microexpression activity =0.2 (low). Through neural networks Calculate the observation interest score =0.60.

[0054] Calculating social cognitive bias: Original deviation: .

[0055] Pressure coupling correction: (The value is very high.)

[0056] Deviation direction: This is an "overreaction".

[0057] Source attribution: System analysis shows that the main stress comes from the U01-U05 group, and they all focus on the abstract drawing A, so it is marked as "group stress-dominated" bias.

[0058] Strategic decision-making: Three-dimensional decision space coordinates: .

[0059] It falls into region A (anti-conformity catalysis zone) because of high bias, high pressure, and positive direction (actual interest is higher than its independent baseline).

[0060] Fuzzy logic confidence Decision generation: Employing an "anti-herd catalysis" strategy, intensity factor .

[0061] Step S400: Adaptive bootstrap content generation and triggering; Content generation (anti-conformity catalysis): Search and Scoring: The system retrieves content related to painting A in the categories of "critical viewpoints" and "comparative analysis" from the knowledge base. One piece of content, c1, received a high score: c1.type = “Comparative Analysis” c1.depth = 6 c1.stance = -0.3 (with a slightly critical tone) c1.target_vector matches "analytical cognitive style". c1.semantic_embedding is related to the critical theory of abstract expressionism.

[0062] The key scoring item, contradiction_strength(c1, S_U01), is very high because the content of c1 points to "the controversy over unconscious brushstrokes in abstract expressionism," which forms a semantic contrast with the general enthusiasm of the audience.

[0063] Variable instantiation and narrative generation: The system traced the social field and found that the main source of influence, U01 and others, had previously gathered at another abstract painting, C.

[0064] Generative spatial narrative framework: "Many viewers, after the pleasant experience of 'Color Variations' (painting C), come to this painting, 'Order of Chaos' (painting A), and continue to seek a strong visual impact. However, art critic X believes that this overemphasis on 'instant emotion' may cause us to overlook the intrinsic value of painting as a craft." Insert a guiding question: "As an architecture enthusiast who is sensitive to structure, do you feel that this creative approach that completely abandons figurative and pre-set structures has a fundamental tension with the 'form follows function' principle that you are familiar with?" Dynamic triggering: Cognitive rhythm analysis: Mr. Lin has been viewing painting A for 90 seconds, and his cognitive load is... Frequency domain analysis shows that it is in a stable "evaluation and thinking" rhythm, cognitive phase Approach a window that is suitable for receiving new information.

[0065] Joint condition determination: Readiness: Sufficient dwell time, stable gaze, value 0.9.

[0066] Receptivity: The current load is moderate, with a value of 0.7.

[0067] Context suitability: The crowd is slightly dispersed, and the noise is reduced, with a value of 0.8.

[0068] Trigger(t) = σ(0.9+0.7+0.8-1.2) ≈ 0.85>0.6.

[0069] Progressive Trigger: The system first displays a gentle question mark icon on the edge of Mr. Lin's AR glasses. He looks at the icon and stays there for 2.5 seconds (active attention). The system then initiates the second-level trigger, playing approximately 5 seconds of introductory audio: "Another perspective on this painting..." Mr. Lin nods slightly (micro-movement detected by millimeter-wave radar). The system determines this as a clear intention to receive the audio and triggers the delivery of the complete spatial narrative audio (approximately 45 seconds).

[0070] Effect evaluation and system evolution: Short-term effect: Within 30 seconds of the audio playback, the system detected Mr. Lin: The gaze begins to move from the center of painting A toward the brushstroke details at the edges (exploration behavior increases).

[0071] social pressure It dropped to 0.7 (the system guided him to psychologically "detach" from the group).

[0072] Observational Interest It decreased slightly from 0.60 to 0.55, getting closer to its baseline of 0.35 (the herd effect was partially corrected).

[0073] He then turned and walked toward painting B (realistic portrait) in the corner.

[0074] Learning feedback: The "effect vector" of this intervention shows that consistency_gain is positive (more consistent cognition). The system records the (s,a,r) tuple for this intervention and uses it to update the reinforcement learning model. In the future, in scenarios similar to "high-density abstract painting exhibition area - audience with architectural background", the tendency to adopt the "anti-conformity catalysis" strategy will increase.

[0075] Through this case, the system is no longer merely a tool for one-way information delivery, but has become a "cognitive partner" capable of sensing the social context, understanding an individual's underlying cognitive biases, and providing a catalyst for dialectical thinking at appropriate psychological junctures. It successfully transformed Mr. Lin's potentially superficial, conformist exhibition visit into a profound experience that triggered personal professional reflection and guided him to explore his true interests (ultimately leading him to painting B). This is precisely the core value of "enhanced cognitive autonomy" that this invention pursues, transcending mere information transmission.

[0076] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for guided art tours based on social influence field modeling, characterized in that, Including the following steps: S1 00: Synchronously perceive the individual visual focus trajectory of all visitors in the exhibition hall, the public behavior representation that can be perceived by others, and the topological characteristics of the exhibition hall environment; S200: Define individual influence sources based on the aforementioned public behavior representations, and construct a dynamic social influence field covering the exhibition hall by combining spatial attenuation and line-of-sight occlusion models; Calculate the environmental social impact stress value of each audience member's location; S300: Maintain an independent cognitive baseline model for each viewer. By comparing the interest score predicted by the model with the real-time observed interest score, and combining the environmental social influence pressure value, calculate the social cognitive bias, and make decisions based on the anti-conformity catalysis, cognitive anchoring reinforcement, or silent observation strategy. S400: Based on the selected strategy, generate and trigger socially restorative guided tour content that includes critical viewpoints, comparative analysis, or in-depth interpretation. The generation or triggering logic of the content is related to the spatial tracing of the main sources of social impact pressure.

2. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, In S200, the formula for calculating the dynamic social influence field strength Φ(x,t) at spatial point x is: in, The intensity of the influence source of the audience u at time t is modulated by its real-time public behavior representation; Let u be the coordinates of the visual focus. Let x be a spatial point and the focus. The line-of-sight obstruction coefficient between them; σ is the influence space attenuation coefficient.

3. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, In S300, social cognitive bias The calculation formula is: in, To score the audience's observed interest at time t, The predicted interest score of the current exhibit i is determined by its independent cognitive baseline model. Let v be the environmental and social stress value experienced by v at time t, and ε be a smoothing constant.

4. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, In S300, when the degree of social cognitive bias Social and environmental stress levels When all values ​​exceed the corresponding threshold, an anti-conformity catalysis strategy is selected to generate questioning or comparative guided content designed to stimulate critical thinking; when... Low but When the level is extremely high, a cognitive anchoring reinforcement strategy is selected to generate in-depth interpretive content aimed at consolidating independent cognition.

5. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, In S400, when generating socially restorative guided tour content, the spatial clusters corresponding to the main sources of influence that constitute the current audience's high environmental social pressure and their historical exhibits are traced back to them. The historical exhibits are then compared with the current exhibits at the content level to construct an implicit narrative that transcends the exhibition space.

6. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, The intensity of micro-responses in the public behavior representation It is calculated by integrating the detection results of exclamatory sounds by lightweight real-time audio analysis and the results of posture change amplitude analysis based on computer vision.

7. The painting guidance method based on social influence field modeling as described in claim 1, characterized in that, The Independent Cognitive Baseline Model It is a personalized interest prediction model trained based on the behavioral data of audience v during periods of low environmental social influence stress in historical exhibition visits.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a painting navigation method based on social influence field modeling as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a painting navigation method based on social influence field modeling as described in any one of claims 1 to 7.

Images