A User Attention Prediction Method Based on Cognitive Models

Abstract

This invention relates to the field of computer system technology based on specific computational models, and discloses a user attention prediction method based on a cognitive model, comprising: dynamically calibrating a user's personalized divergence rate benchmark at the initial stage of a session, and generating a discrimination threshold accordingly; then continuously calculating the spatial structural entropy of content elements on the canvas and the behavioral rhythm of user operations, and identifying the user's cognitive state transition from divergence to convergence based on the specific synergistic relationship presented by the respective time change rates of these two physical quantities; further, diagnosing the existence of cognitive deadlock by examining the causal relationship between the amount of organizational behavior input and the actual entropy reduction effect. This invention establishes the judgment of the user's internal thinking pattern on a quantifiable dynamic system evolution process, and has the ability to evaluate the effectiveness of the cognitive process, and can distinguish between productive organization and ineffective struggle, two situations with similar external behaviors but completely different internal cognitive states.

Description

Technical Field

[0001] This invention relates to a user attention prediction method based on a cognitive model, belonging to the technical field of computer systems based on specific computational models. Background Technology

[0002] In complex intellectual activities such as collaborative design, strategic planning, or knowledge building, collaborative tools like digital whiteboards have become indispensable mediums. Their core task is to support and facilitate users' transition from divergent information listing to convergent, structured insights. Currently, the understanding and analysis of user attention in such tools generally relies on capturing the physical trajectory of user actions, such as recording the click frequency of mouse movement hotspots or eye-tracking fixation points. These methods are effective in depicting the instantaneous physical focus of user actions.

[0003] However, when we examine these technologies within a real creative process—from chaotic exploration to orderly integration—their inherent limitations arise from the core assumption that physical location equates to cognitive focus. These limitations manifest in several ways: In a typical group brainstorming scenario, users may rapidly create numerous seemingly unrelated fragments of information across the canvas during the divergent phase. At this point, existing technologies, besides concluding that the user's attention is highly scattered, cannot interpret the underlying active exploratory intent. More importantly, when the process enters the convergence phase, and users begin frequently moving, connecting, and reorganizing elements, although their physical operational trajectories may be even more chaotic and intense than in the divergent phase, an orderly cognitive structure is emerging from the information. Faced with this orderly chaos, existing technologies, unable to discern the cognitive state shifts behind physical behavior, cannot identify these shifts. They can record what the user did, but cannot know which step the user has reached.

[0004] To address this issue, the industry has attempted to infer user intent through more complex behavioral sequence pattern matching. However, this approach often falls into an infinite enumeration of behavioral patterns, which not only incurs a huge computational burden but also fails to fundamentally break free from dependence on isolated physical events. It still cannot answer a deeper question: how to identify the emergence of macroscopic cognitive phase transitions from a series of seemingly unrelated micro-behaviors. Specifically, existing technologies suffer from the following shortcomings: 1. They treat user behavioral data as a sequence of physical events without temporal context, lacking a mechanism for interpretation from the perspective of the overall system evolution; 2. They confuse the physical trajectory of attention with cognitive intent, failing to distinguish between efficient structured thinking and intensive ineffective organization; 3. They lack a simple, universal computational framework that is independent of specific task models to quantify and predict this cognitive state transition from disorder to order. Therefore, the technical problem to be solved by this invention is how to establish a new way of perception that no longer simply tracks the physical coordinates of user behavior, but can accurately perceive whether the user's cognitive process has undergone a key nonlinear state transition from divergent exploration to convergent integration by analyzing the dynamic synergistic relationship between the overall layout evolution of information elements on the canvas and the rhythm of user operation, and further understand the effectiveness of the process. Summary of the Invention

[0005] This invention provides a user attention prediction method based on a cognitive model. Its main purpose is to solve the problem that existing technologies cannot accurately perceive whether a user's cognitive process has undergone a state transition from divergent exploration to convergent integration by analyzing the user's operational behavior on a digital canvas, and further gain insight into the effectiveness of the process.

[0006] To achieve the above objectives, the present invention provides a user attention prediction method based on a cognitive model, comprising:

[0007] Within the initial time window of a session in a digital whiteboard system, when a behavioral rhythm value representing the type of user operation is consistently below a creative behavior dominance threshold, a baseline value is measured and determined to characterize the user's personalized divergence rate.

[0008] Obtain the spatial coordinates of the content elements on the canvas to calculate the spatial structure entropy value, and obtain user operation events to calculate the behavior rhythm value;

[0009] Calculate the time rate of change of spatial structure entropy and the time rate of change of behavioral rhythm values;

[0010] When the time change rate of the spatial structure entropy is less than a first negative threshold dynamically generated based on a personalized divergence rate benchmark, and the time change rate of the behavioral rhythm value is greater than a second positive threshold, a switching marker representing the user's cognitive state switching from a divergent state to a convergent state is generated.

[0011] After generating the switching flag, the expected reduction of spatial structure entropy is determined based on the amount of input of organizational behavior, and when the actual reduction of spatial structure entropy is consistently lower than the expected reduction within a continuous time window, an attention state signal representing a convergent-stalemate state is generated.

[0012] Preferably, the calculation of spatial structure entropy involves obtaining the coordinates of the center point of all content elements on the canvas, calculating the common geometric centroid of all content elements, and further calculating the variance of the Euclidean distance from the center point of each content element to the geometric centroid; the calculation of behavioral rhythm value involves dividing user operation events into creation events and organization events, and calculating the proportion of the number of organization events to the total number of creation events and organization events within the sliding time window.

[0013] Preferably, the step of dynamically generating the first negative threshold specifically includes: within the initial time window of the session, under the premise that the verification behavior rhythm value is continuously lower than the creation behavior dominance threshold, continuously measuring the time change rate of the spatial structure entropy value, and calculating its time average as the personalized divergence rate benchmark value. ; and in accordance with the rules Generate the absolute value of the first negative threshold It is a proportionality coefficient whose value ranges from 0.1 to 0.9.

[0014] Preferably, the input amount of organizational behaviors is calculated by weighting and summing the organizational behaviors according to their types within a time window; the expected reduction of spatial structure entropy is determined by multiplying the input amount of organizational behaviors by an entropy transformation coefficient; when the actual reduction of spatial structure entropy is lower than the expected reduction amount for three consecutive time windows, an attention state signal of convergence-stalemate is generated.

[0015] Preferably, the method further includes: after generating the attention state signal of the convergence-stalemate state, obtaining the location coordinates of all organizational behaviors on the canvas during the stalemate determination period; calculating the spatial distribution variance of the location coordinates of all organizational behaviors; and comparing the spatial distribution variance with one or more diagnostic thresholds to generate a classification signal that further distinguishes the convergence-stalemate state into a locally fixed stalemate state or a globally disordered stalemate state.

[0016] Preferably, the method further includes: after generating the classification signal, continuously performing the distinction between convergence-stalemate states within a subsequent sliding time window to generate a state time series of stalemate type; and monitoring the state time series, and adaptively adjusting the auxiliary intervention strategy for the user when the stalemate type transitions from a locally fixed type to a globally disordered type.

[0017] Preferably, the method further includes: determining a complexity factor that characterizes the total number of content elements on the canvas or the global spatial structure entropy value before calculating the spatial distribution variance of the location coordinates of all organizational behaviors; and the diagnostic threshold used to distinguish between local fixed deadlock states and global disordered deadlock states is dynamically normalized based on the complexity factor through a logarithmic function.

[0018] Preferably, the method further includes: when the spatial structure entropy value is continuously higher than a chaotic state threshold, initiating and monitoring the duration of silence when the user has no creative or organizational behaviors; when the duration of silence exceeds a paralysis tolerance duration, generating a representation analysis-paralysis state attention signal.

[0019] Preferably, the calculation of the time change rate of spatial structure entropy and the time change rate of behavioral rhythm is obtained by performing first-order difference operations on the spatial structure entropy sequence and the behavioral rhythm sequence respectively, and then performing exponential smoothing filtering on the difference results; the length of the sliding time window is set to ten to sixty seconds, and the sampling frequency of user operation events is not less than ten hertz.

[0020] Preferably, the method further includes: executing a corresponding whiteboard interaction strategy based on the generated classification signal or the attention state signal of the analysis-paralysis state; the interaction strategy includes highlighting untouched external element clusters on the canvas for local fixed deadlock states, temporarily hiding non-core canvas modules for global disordered deadlock states, and providing a simplified starter template for the analysis-paralysis state.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. This invention first acquires the spatial location information of content elements on a digital canvas and the sequence of user operation events. It does not analyze the static structure of the element layout or a single classification of user behavior in isolation. Instead, it calculates the spatial structure entropy value, which represents the degree of disorder in the layout, and the temporal behavior rhythm value, which represents the behavior pattern, and continuously monitors the time change rate of each of these two physical quantities. Only when a specific dynamic coupling relationship occurs, such as a significant decrease in the time change rate of the spatial structure entropy and a significant increase in the time change rate of the temporal behavior rhythm, is it determined that a cognitive state transition from divergent exploration to convergent integration has occurred. This approach establishes the judgment of the user's internal thinking pattern on an observable and quantifiable physical system evolution process, so that the prediction of attention is no longer based on static rule matching of specific behavioral events, but on a dynamic system phase transition for identification.

[0023] 2. After determining that the user has entered a convergence and integration state, this invention further establishes an evaluation mechanism for the effectiveness of the cognitive process. It no longer simply accepts the fact that the user is organizing, but continuously examines the correlation between the amount of the user's organizational behavior and the actual reduction in spatial structural entropy. When there is a persistent mismatch between the expected causal relationship between the investment of organizational behavior and the resulting improvement in structural orderliness, the system identifies a cognitive deadlock state in which the user intends to converge but the process is blocked. This mechanism enables the system to have a perceptual depth from recognizing the user's intention to evaluating the effectiveness of the process, and can distinguish between effective organization and ineffective struggle, which are similar in external behavior but have completely different internal cognitive states.

[0024] 3. This invention also establishes a multi-level, logically progressive adaptive analysis system. In the initial stage of the conversation, it utilizes the user's purely divergent behavior to measure and establish a personalized divergent rate benchmark specific to that user, which is used to calibrate the judgment threshold for subsequent cognitive state transitions, avoiding the applicability bias of fixed parameters among different users. After identifying cognitive deadlock, it reuses the spatial location information of those organizational behaviors that led to the deadlock, and further distinguishes whether the deadlock stems from excessive fixation in a local area or from global disorder by analyzing the spatial distribution concentration of these behaviors. This closed-loop information flow from personalized calibration to state recognition and then to fault diagnosis and characterization makes each link of the attention prediction method interconnected and work collaboratively. Attached Figure Description

[0025] Figure 1 This is a logical flowchart of the cognitive state evolution and deadlock diagnosis of the present invention;

[0026] Figure 2 This is a functional module and strategy mapping diagram of the attention prediction system of the present invention;

[0027] Figure 3 This is a dynamic timing interaction diagram of multi-module collaborative computing in this invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] This invention provides a user attention prediction method based on a cognitive model, comprising an online personalized calibration step, a continuous cognitive state monitoring step, and a cognitive process effectiveness diagnosis step. The method operates on directly accessible data streams within a digital whiteboard system. Its core technical solution lies in transforming the inference of user psychological states into an objective analysis of the state evolution of an observable and quantifiable computational model system. In a collaborative whiteboard application, different users have different operating rhythms and thinking habits. Using fixed parameter thresholds to judge cognitive states can easily lead to oversensitivity for some users and undersensitivity for others. To address this challenge, the method of this invention is configured to first perform adaptive calibration of the user's personalized cognitive rhythm. Specifically, at the start of a new whiteboard session, the system defines an initial time window, such as the first sixty seconds after the session begins, and continuously monitors a behavioral rhythm value representing the user's operation type within this window. When the If the value remains below a threshold indicating dominant creative behavior, such as 0.1, the system confirms that the current user is in a divergent exploration phase dominated by creative behaviors. Once this condition is met, the system continuously samples and calculates the canvas space structure entropy value at a frequency of no less than 10 Hz within this initial time window. rate of change over time Furthermore, the time average of this rate of change within this window is calculated to obtain a personalized divergence rate benchmark value specific to this user's current session. Subsequently, the system is based on this. Value, through rules To dynamically generate the absolute value of the first negative threshold used for subsequent state determination The proportionality coefficient This is a settable parameter with a value ranging from 0.1 to 0.9, which reflects the sensitivity of the system's judgment. For example, a default value such as 0.5 can be set. This step aims to adjust the judgment scale used in subsequent cognitive phase transition recognition to adapt it to the individual user's behavior pattern.

[0030] After completing personalized threshold calibration, in order to identify the user's cognitive state transition from divergent exploration to convergent integration, this method enters a continuous monitoring phase based on spatiotemporal feature collaborative perception. In this phase, two dimensions of information flow are processed in parallel, the first dimension being spatial structure entropy. The quantization process involves the system acquiring the center point coordinates of all N independent content elements on the canvas in real time. Calculate the common geometric centroid of all content elements. Furthermore, the variance of the Euclidean distance from the center point of each content element to the geometric centroid is calculated, and this variance value is used as... The value; the second dimension is the temporal rhythm of behavior. The representation of this is determined by the following calculation procedure: the system divides user operation events into creation events and organization events, and within a sliding time window with a length that can be set to ten to sixty seconds, calculates the proportion of organization events to the total number of creation and organization events within that window. This proportion is the [representation / statistic]. The value; then, the system analyzes the collected data... Value sequence and The value sequences are subjected to first-order difference operations, and the difference results are then subjected to exponential smoothing filtering to obtain their respective rates of change over time. and The determination of a cognitive state transition is based on a specific synergistic relationship formed by the changing trends of two dimensions; that is, when the system detects... Less than the aforementioned personalized divergence rate benchmark value Dynamically generated first negative threshold ,and Greater than a second positive threshold Only when the system generates a switching flag representing the user's cognitive state switching from a divergent state to a convergent state.

[0031] In real workflows, even if users intend to converge, their organizational behavior may be inefficient. To evaluate the effectiveness of such processes, the method of this invention activates a diagnostic mechanism for the effectiveness of a cognitive process after generating the aforementioned switching marker. This mechanism aims to examine whether a causal relationship exists between organizational behavior input and the resulting improvement in structural orderliness. Specifically, the implementation procedure is as follows: First, within a time window, the system performs a weighted summation of all organizational behaviors that have occurred according to their type to calculate the amount of organizational behavior input. Secondly, the system calculates the amount of organizational behavior input. Multiply by an entropy transformation to calculate the coefficients To determine the expected reduction in the entropy of a spatial structure. Among them, entropy transformation calculates coefficients. It is an empirical constant obtained through offline data analysis for calibrating organizational efficiency. Its physical meaning is the average entropy reduction that can be achieved by a unit of organizational behavioral input. Ultimately, the system continuously compares the actual reduction in spatial structure entropy within consecutive time windows. Compared with the expected reduction When detected Below in three consecutive time windows At this time, the system generates an attention state signal that represents a convergence-stalemate state. Through this mechanism, the system can distinguish between productive organizing behavior and ineffective struggling behavior.

[0032] To further differentiate between types of cognitive deadlock, after generating the attention state signal of the convergence-dead state, the system activates a deadlock type diagnosis and tracking module. This module first obtains the location coordinates of all organizational behaviors on the canvas during the deadlock determination period and calculates the spatial distribution variance of these coordinate points. To eliminate potential misjudgments that might occur with a fixed threshold on canvases of varying complexity, the system calculates... Previously, a contextual scale factor representing the current global complexity of the canvas would be determined first, such as the total number of content elements. Subsequently, the diagnostic threshold used to distinguish the type of deadlock is based on this complexity factor. Dynamic normalization is performed using a logarithmic function to generate a value that dynamically changes with task complexity. and The system will calculate Compared with this pair of dynamic thresholds, if This generates a classification signal that further distinguishes convergent-stalemate states into locally fixed stalemate states. Then, a classification signal is generated to distinguish between global disordered deadlock states; this diagnostic distinction action will be continuously executed within a subsequent sliding time window to generate a state time series of deadlock types. When the system detects a state transition in this series, it adaptively adjusts the auxiliary intervention strategy for the user; to cover passive deadlock scenarios where users are inactive for a long time due to information overload, the method of this invention also includes an analysis-paralysis state recognition mechanism; the procedure of this mechanism is as follows: when the system detects the spatial structure entropy value Persistently above a chaotic state threshold At that time, the system will start and monitor the duration of the user's silent behavior when there is no creative or organizational behavior. When the More than one paralysis tolerance period For example, at 90 seconds, the system generates a representation analysis-paralyzed attention state signal; this mechanism is combined with the aforementioned mechanism for handling active deadlock to cover both active and passive cognitive dilemmas that users may encounter during the convergence phase.

[0033] Example 1: In an online collaborative scenario involving members from multiple fields designing a complex system architecture, the digital whiteboard is initially filled with hundreds of independent content elements representing the functional requirements and interface protocols of technical modules. At this point, the system calculates the spatial structure entropy value. In a high-level position, team members primarily engage in creative activities, with a behavioral rhythm value of [missing information]. The value continues to approach zero; in this initial stage, the system measures... The time-varying rate of change was used to determine a personalized divergence rate baseline for the team's collaborative sessions. Based on this, a dynamic judgment threshold was generated. As the discussion deepened, the team began to attempt to restructure the scattered modules. User behavior shifted from creating new elements to moving and connecting existing elements extensively. During this stage, the canvas exhibited drastic and widespread changes in element positions at the physical operation level. However, analysis based on the physical trajectory of user behavior would be interpreted as continuous chaos because it could not identify the cognitive intent behind the behavior. The computational model deployed in this scenario monitored the spatial structure entropy in parallel. With time-related behavioral rhythms The rate of change of these two physical quantities over time, when they are calculated The rate of change over time is less than that through Dynamically generated negative threshold ,and The rate of change over time is greater than a second positive threshold. At that time, the combination of this set of spatiotemporal evolution features enabled the system to generate a switching marker from the divergent state to the convergent state. This recognition process does not rely on the analysis of the semantics of user operations, but is based on the synchronous changes of two physical quantities: the orderliness of information layout and the proportion of organizational behavior.

[0034] After the system determined that the team had entered a consolidation state, the team encountered difficulties in organizing a group of core computing modules. Members repeatedly dragged several elements within the group and tried different connection relationships, but the overall architecture was not optimized. At this point, although the team's organizational behaviors were frequent, the overall orderliness was not maintained. The value decreased slowly over several consecutive time windows; the system then activated the cognitive process effectiveness diagnostic mechanism, which calculated the higher organizational behavioral input during this period. Based on this, an expected reduction in spatial structure entropy was determined. Due to the actual reduction in entropy If the value remains consistently below the expected value, the system generates an attention state signal indicating a convergent-stalemate state. Subsequently, a stalemate type diagnostic module begins operation. This module reuses the location coordinates of the organizational behaviors that led to the stalemate signal to calculate the spatial distribution variance of these coordinate points. The value is very small, then it depends on the total number of content elements on the current canvas. After dynamically normalizing and adjusting the diagnostic threshold, the system determines that... The value is less than the adjusted value. This generates a classification signal that further distinguishes the deadlock state into a locally fixed deadlock state. Based on this classification signal, the system executes the corresponding whiteboard interaction strategy, highlighting an external element cluster on the canvas that was created early but has not been touched for a long time and is related to the current fixed module. The team's work then enters a new structured stage, and the new system architecture gradually takes shape on the canvas.

[0035] Example 2: To distinguish between scenarios with similar external behaviors but different internal cognitive states, this invention conducted a comparative experiment. The experimental platform was based on a standardized digital whiteboard software environment. A user behavior simulation script agent executed a series of preset task sequences with clearly defined cognitive states to ensure the reproducibility of the experimental conditions. The sliding time window length in the experiment was set to 30 seconds, based on the balance between the real-time performance of state recognition and the stability of data smoothness. The data sampling frequency was set to 10 Hz to capture all discrete user interaction events. The experiment set up a control group and an experimental group. The control group adopted a conventional analysis method based on the frequency of user operation events, which output a quantitative index, namely the total number of user operations per unit time. The experimental group deployed the complete calculation model of this invention. In the experiment, the output of the control group, namely the total number of operations per unit time, was recorded as 115, 110, and 125 in the efficient convergence scenario S2 (locally fixed deadlock scenario S3) and the globally disordered deadlock scenario S4, respectively. The values ​​were all high and there was no significant difference. This shows that the analysis method based solely on the operation frequency cannot distinguish between efficient organizational behavior and two different types of ineffective struggle behavior.

[0036] In contrast, the computational model deployed in the experimental group effectively distinguished between the aforementioned scenarios. In scenario S2, the model correctly output the convergence state signal; while in scenarios S3 and S4 with similar operands, the model generated convergence-stalemate attention state signals. This distinction was achieved by the model continuously increasing the user's organizational behavior input after determining that the user had entered the convergence state. The actual reduction in spatial structure entropy resulting from this A causal relationship test was performed in scenarios S3 and S4, because... and A persistent mismatch occurred, and the system identified a cognitive deadlock state. Further differentiation between S3 and S4 was achieved by analyzing the spatial distribution variance of the organizational behaviors that led to the deadlock. This allows the two to be identified as locally fixed and globally disordered, respectively. In the analysis paralysis scenario S5, where the user has not performed any operation for a long time, the experimental group also generated an analysis-paralysis state signal by detecting high-entropy silence and recording 0 operation counts, while the control group had no effective output in this scenario. The experimental results show that the method of the present invention, by constructing a multi-level computing model, can identify and distinguish various user behavior sequences that appear indistinguishable under conventional analysis methods. In particular, it can distinguish between productive organization and ineffective struggle, laying a data foundation for providing interactive assistance synchronized with the user's cognitive state.

[0037] Example 3: This example combines Figures 1 to 3 This section describes a user attention prediction method based on a cognitive model, such as... Figure 1 As shown, the diagram begins with the step of acquiring user input, namely obtaining the canvas space coordinates and the sequence of user operation events, and from this, two processing paths unfold in parallel; one is an offline step of dynamically calibrating a personalized divergence rate benchmark, by calculating user-specific... Values ​​are used to generate a judgment threshold. The second is a continuous online monitoring cycle, which first calculates the spatial structure entropy. Calculating behavioral rhythm values This involves calculating the rate of change over time to capture the dynamic evolution of entropy and rhythm; the core step in determining cognitive state transitions is to examine the rate of change of entropy over time. Is it less than the negative threshold? And the temporal rate of change of rhythm Is it greater than the positive threshold? This collaborative relationship generates a switching marker to identify the transition from a divergent state to a convergent state. After entering the convergent state, the system continuously compares the actual reduction of entropy with the expected value through a diagnostic cognitive deadlock module. If the entropy reduction is consistently lower than the expected value, a convergence-deadlock state signal is generated. Otherwise, the system identifies the state as efficient convergence and continuously monitors it. For deadlock states, the system analyzes the spatial distribution variance of organizational behavior through a key module: deadlock type diagnosis, further classifying it into local fixed deadlock or global disordered deadlock. In addition, a parallel analysis-paralysis identification module generates an analysis paralysis state signal by monitoring long-term inactive behavior in a high-entropy state. All the finally identified state signals jointly guide the output of attention signals and the execution of the corresponding whiteboard interaction strategy in the final stage.

[0038] like Figure 2As shown, in this architecture, the top-level digital whiteboard system provides data input for three core modules. The personalized calibration module is responsible for monitoring the initial time window and calculating the divergence rate benchmark value, and performs dynamic threshold generation accordingly to produce the first negative threshold for state determination. The cognitive state monitoring module performs parallel spatial structure entropy calculation and behavioral rhythm value calculation, and obtains the time change rate through first-order difference + smoothing filtering. The process effectiveness diagnosis module is activated after recognizing the convergence state. It contains sub-modules for analyzing organizational behavior input and verifying entropy reduction effect to identify deadlocks. The deadlock type diagnosis sub-module performs qualitative analysis of the deadlock through spatial distribution variance analysis. These three modules work together to output a series of cognitive state judgment results, such as divergent exploration state, convergent integration state, locally fixed deadlock, globally disordered deadlock, and analysis paralysis state. Ultimately, it drives the underlying adaptive interaction strategy, such as highlighting untouched elements, hiding non-core modules, providing starting templates, or dynamically adjusting parameters for different states.

[0039] like Figure 3 As shown in the diagram, this depicts a parallel information processing process that begins with the user's continuous interaction with the canvas. On one hand, the whiteboard system acquires element coordinates in real time and transmits them to the entropy monitoring module, which sequentially performs operations to calculate the geometric centroid, spatial structure entropy, and the entropy time change rate. On the other hand, the whiteboard system synchronously captures user operation events and transmits them to the behavior analysis module, which then performs operations to distinguish between creation / organization events, calculate the behavior rhythm value, and calculate the rhythm time change rate. The entropy monitoring module and the behavior analysis module send their respective calculated change rate results—the negative threshold for entropy change rate and the positive threshold for rhythm change rate—to the state determination module. This module verifies the collaborative relationship to determine whether a state transition exists and generates a state switching flag when the conditions are met. Finally, it feeds back the signal indicating a divergence-to-convergence transition to the upper-layer application or the user.

[0040] Example 4: When deploying the computational model of this invention in collaborative tasks of different natures, such as highly structured circuit diagram design or low-structured free creative brainstorming, a fixed entropy transformation coefficient is used. A fixed set of deadlock diagnosis thresholds may lead to decreased model accuracy due to its inability to adapt to the differences in the relationship between user behavior and canvas information evolution in different tasks. To address this issue, this invention provides an offline, systematic parameter calibration procedure. This procedure first establishes a template library containing various collaborative task types, including software architecture design business model canvases and free divergent discussions. Then, a user behavior simulation script agent is used to execute multiple pre-defined behavioral sequences with clearly defined cognitive state labels under each task template, including sequences labeled as efficiently convergent and sequences labeled as locally fixed, and records all relevant computational model input and output data during this process. For entropy transformation coefficients... The system is calibrated so that it only analyzes the data generated by efficient convergent behavioral sequences and records the organizational behavioral input within them. The actual reduction in sequence and spatial structure entropy The sequence undergoes linear regression analysis. The slope of the regression curve, in physical terms, represents the average entropy reduction resulting from a unit of organizational behavioral input under that specific task type. Therefore, it is determined as the entropy transformation coefficient corresponding to that task type. The value of .

[0041] For dynamic threshold functions used to distinguish deadlock types and The parameters in the system follow similar calibration procedures, but the data dimensions analyzed differ. The system extracts two types of behavioral sequences—local fixation and global disorder—at different canvas complexities, i.e., the total number of content elements. Variance of spatial distribution of organizational behavior The data points are plotted on a graph. The x-axis is... On a two-dimensional plane with the vertical axis as the ordinate, two separable data clusters are formed. Then, by using support vector machines or other statistical partitioning methods, a boundary function that can separate these two data clusters is calculated, and this boundary function or its offset function is used as the final dynamic threshold function to distinguish between local fixed deadlocks and global disordered deadlocks. By executing this set of offline calibration procedures, the system can generate a library of validated parameter sets for different task types. During actual operation, the system can load the corresponding parameter set according to the current task type, or use a general average parameter set generated from all template data when the task type is unknown, thereby providing appropriate parameter configurations for different types of collaborative tasks.

[0042] Example 5: In a collaborative session initiated from a preset template, because the initial state of the canvas contains a large number of content elements, users may directly begin organizing behaviors within the initial time window of the session, resulting in the inability to meet the personalized divergence rate benchmark. The prerequisite for measurement is the behavioral rhythm value. If the content remains below the threshold for dominant creative behavior, the system executes a set of pre-context recognition and parameter loading procedures. When the session starts, the system first analyzes the initial content features of the canvas. If the canvas is empty, it loads a set of baseline parameters for general divergent task calibration. If the canvas content conforms to the structural features of a specific template, it loads a set of specific parameters that correspond to it and have been generated through offline calibration.

[0043] This parameter set contains entropy transformation coefficients applicable to this task type. A dynamic threshold function for deadlock diagnosis, and a non-personalized convergence decision threshold for backup. When the system detects that the preset session initial time window has ended, the personalized divergence rate benchmark... If the measurement conditions are still not met, the system skips the personalized threshold generation step and directly uses the preset parameters in the parameter set in subsequent operations. This serves as the first negative threshold used to determine the convergence state.

[0044] Example 6: In a long-term collaborative project lasting several hours or even days, the nature of the task evolves, from initial divergent information gathering to mid-term structured framework building, and then to later refined content correction. If the computational model uses the same set of fixed parameters loaded at the beginning of the session throughout the entire process, its prediction accuracy will decrease as the nature of the task deviates. To address this situation, the method of this invention may include an online parameter adaptive fine-tuning procedure. This procedure performs regression analysis on recent historical data in the background at a preset time period, such as every fifteen minutes. Specifically, the system extracts all time segments identified as having efficient convergence within this period and allocates resources to the organizational activities within these segments. The actual reduction in sequence and spatial structure entropy The sequence is subjected to linear regression again to obtain a local entropy transformation coefficient that reflects recent working characteristics. If the Values ​​and coefficients currently being used by the model If the deviation of the value exceeds a preset adjustment trigger threshold, the system determines that the nature of the task has changed and accordingly adjusts the current coefficient value accordingly. The values ​​are adjusted for a smooth transition.

[0045] Correspondingly, the smoothing coefficient in the exponential smoothing filter used to calculate the rate of change over time... And a second positive threshold used to determine the convergence state. This process also involves collaborative updates within the fine-tuning procedure; smoothing coefficient The value of is related to the signal-to-noise ratio of the recent spatial structure entropy sequence. During a phase of rapid addition and deletion of content elements, the signal-to-noise ratio is relatively high, and the system uses a smaller . A larger value is used to enhance the smoothing effect, while a larger value is used during a stable consolidation phase. Values ​​to improve response speed; second positive threshold The value of this parameter is related to the total frequency of recent user operations. During high-frequency operation phases, to avoid misjudgments caused by occasional behavioral fluctuations, The value of is appropriately increased; through this online adaptive fine-tuning procedure, the core parameters of the computational model thus maintain a dynamic match with the current work content in long-cycle tasks. In another application scenario, when the computational model determines that the user's state is a globally disordered deadlock, it is based on the observation that the user's organizational behavior is highly discrete in space, i.e. The value is greater than the value after normalization by the complexity factor. At this point, the system's interaction strategy is not to highlight specific elements, but rather to temporarily and semi-transparently hide some secondary, non-core interface modules on the canvas, such as toolbars or comment panels. This reduces the visual complexity of the interface and helps users refocus their attention on the core content elements. In the scenario, when the system detects that the user has exceeded the preset paralysis tolerance time in the high-entropy state at the beginning of the session... Without performing any operations and generating an analysis-paralyzed attention state signal, the system will proactively provide a structured, simplified starting template in the central area of ​​the canvas to guide thinking. This template can be a two-row, two-column matrix or a basic central radiating diagram to lower the initial cognitive threshold for users to start organizing and tidying up.

[0046] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A user attention prediction method based on a cognitive model, characterized in that, include: Within the initial time window of a session in a digital whiteboard system, when a behavioral rhythm value representing the type of user operation is consistently below a creative behavior dominance threshold, a baseline value is measured and determined to characterize the user's personalized divergence rate. Obtain the spatial coordinates of the content elements on the canvas to calculate the spatial structure entropy value, and obtain user operation events to calculate the behavior rhythm value; Calculate the time rate of change of spatial structure entropy and the time rate of change of behavioral rhythm values; When the time change rate of the spatial structure entropy is less than a first negative threshold dynamically generated based on a personalized divergence rate benchmark, and the time change rate of the behavioral rhythm value is greater than a second positive threshold, a switching marker representing the user's cognitive state switching from a divergent state to a convergent state is generated. After generating the switching flag, the expected reduction of spatial structure entropy is determined based on the amount of input of organizational behavior, and when the actual reduction of spatial structure entropy is consistently lower than the expected reduction within a continuous time window, an attention state signal of convergence-stalemate is generated. Specifically, the calculation of spatial structure entropy involves obtaining the coordinates of the center point of all content elements on the canvas, calculating the common geometric centroid of all content elements, and further calculating the variance of the Euclidean distance from the center point of each content element to the geometric centroid; the calculation of behavioral rhythm value involves dividing user operation events into creation events and organization events, and calculating the proportion of the number of organization events to the total number of creation events and organization events within the sliding time window. The steps for dynamically generating the first negative threshold specifically include: within the initial time window of the session, under the premise that the verified behavioral rhythm value is continuously lower than the threshold for dominant creative behavior, continuously measuring the temporal change rate of the spatial structure entropy value, and calculating its temporal average as the benchmark value for personalized divergence rate. ; and in accordance with the rules Generate the absolute value of the first negative threshold It is a proportionality coefficient whose value ranges from 0.1 to 0.

9.

2. The user attention prediction method based on a cognitive model according to claim 1, characterized in that, The calculation of the input amount of organizational behaviors is obtained by weighting and summing the organizational behaviors according to their types within a time window; the determination of the expected reduction of spatial structure entropy is obtained by multiplying the input amount of organizational behaviors by an entropy transformation coefficient; when the actual reduction of spatial structure entropy is lower than the expected reduction amount for three consecutive time windows, an attention state signal of convergence-stalemate is generated.

3. The user attention prediction method based on a cognitive model according to claim 2, characterized in that, The method further includes: after generating the attention state signal of the convergence-stalemate state, obtaining the location coordinates of all organizational behaviors on the canvas during the stalemate determination period; calculating the spatial distribution variance of the location coordinates of all organizational behaviors; and comparing the spatial distribution variance with one or more diagnostic thresholds to generate a classification signal that further distinguishes the convergence-stalemate state into a locally fixed stalemate state or a globally disordered stalemate state.

4. The user attention prediction method based on a cognitive model according to claim 3, characterized in that, The method also includes: after generating the classification signal, continuously performing the distinction between convergence-stalemate states within a subsequent sliding time window to generate a state time series of stalemate type; and monitoring the state time series, and adaptively adjusting the auxiliary intervention strategy for the user when the stalemate type changes from a locally fixed type to a globally disordered type.

5. The user attention prediction method based on a cognitive model according to claim 3, characterized in that, The method also includes: determining a complexity factor that characterizes the total number of content elements on the canvas or the global spatial structure entropy value before calculating the spatial distribution variance of the location coordinates of all organizational behaviors; and the diagnostic threshold used to distinguish between local fixed deadlock states and global disordered deadlock states is dynamically normalized based on the complexity factor through a logarithmic function.

6. The user attention prediction method based on a cognitive model according to claim 1, characterized in that, The method also includes: when the spatial structure entropy value is continuously higher than a chaotic state threshold, initiating and monitoring the duration of silence when the user has no creative or organizational behaviors; when the duration of silence exceeds a paralysis tolerance duration, generating a representation analysis-paralysis state attention signal.

7. The user attention prediction method based on a cognitive model according to claim 1, characterized in that, The calculation of the time change rate of spatial structure entropy and the time change rate of behavioral rhythm is obtained by performing first-order difference operations on the spatial structure entropy sequence and the behavioral rhythm sequence respectively, and then performing exponential smoothing filtering on the difference results; the length of the sliding time window is set to ten to sixty seconds, and the sampling frequency of user operation events is not less than ten hertz.

8. The user attention prediction method based on a cognitive model according to claim 3, characterized in that, The method also includes: executing a corresponding whiteboard interaction strategy based on the generated classification signal or the attention state signal of the analysis-paralysis state; the interaction strategy includes highlighting untouched external element clusters on the canvas for local fixed deadlock states, temporarily hiding non-core canvas modules for global disordered deadlock states, and providing a simplified starter template for the analysis-paralysis state.

Images