Method, system and electronic device for dynamic cognitive path modeling and prediction, storage medium

Abstract

The present disclosure provides a dynamic cognitive path modeling and prediction method, system, electronic device and storage medium; the method comprises the following steps: collecting learning interface interaction behavior of a learner in real time to generate an original interaction sequence containing multiple time-stamped ordered event tuples; streaming analysis is performed according to a preset behavior analysis rule library, the event tuples are aggregated / converted into semantic micro behaviors with cognitive psychology meaning to obtain a semantic micro behavior sequence; the semantic micro behavior sequence is input into a sequence modeling network to generate a dynamic cognitive path vector encoding the current complete interaction history at each time step; based on the vector, downstream tasks are processed through a prediction head to output real-time intervention signals or cognitive diagnosis reports. The technical scheme of the present disclosure models the fine-grained behavior of the learner during the problem-solving process, realizes dynamic tracking and forward-looking prediction of the cognitive state of the learner, and realizes active and timely personalized intervention.

Description

Technical Field

[0001] This disclosure belongs to the field of artificial intelligence education technology, and in particular relates to a method, system, electronic device, and storage medium for dynamic cognitive path modeling and prediction. Background Technology

[0002] In the field of intelligent education, achieving true personalized learning has always been a core objective, namely, providing individualized instruction based on learners' cognitive levels, learning styles, and thinking patterns. However, existing technologies have consistently failed to overcome a key bottleneck: learners' cognitive processes are treated as a "black box," only observable through inputs (e.g., learning materials) and outputs (e.g., test results), lacking an effective understanding of the complex internal thought processes.

[0003] Existing technologies have evolved primarily along two paths. The first path is based on knowledge state modeling using statistical inference, such as Bayesian knowledge tracing and cognitive diagnostic models. This simplifies learners' knowledge mastery status into binary latent variables (e.g., mastery / not mastery) and relies on answer results (e.g., correct / incorrect) for inference. This approach is inherently static and result-driven, compressing the rich problem-solving process into a single bit of information. It cannot distinguish between different thought processes (e.g., quick problem-solving versus trial-and-error guessing), leading to inaccurate teaching recommendations. The second path is based on sequence data modeling using deep learning, such as deep knowledge tracing. This utilizes recurrent neural networks to process historical answer sequences. While it improves prediction accuracy, the input data is still limited to a binary pair of question ID and answer result, failing to capture the micro-cognitive dynamics during the problem-solving process—like observing the door of a black box rather than its interior.

[0004] In the process of developing the existing technology, the inventors discovered at least the following technical problems: First, it is impossible to analyze how students think and the root causes of errors, such as conceptual confusion or calculation mistakes, which makes the recommended exercises lack specificity; second, the feedback mechanism relies on the submission of answers after the fact, missing the opportunity for real-time intervention when thinking gets stuck; finally, the existing model ignores process-based cognitive data. Summary of the Invention

[0005] The purpose of this disclosure is to address the black box problem of cognitive processes in existing learner models that rely on outcome data, as mentioned in the prior art, and to provide a method, system, and related products that can perform fine-grained, dynamic, and context-aware modeling and prediction of learners' cognitive processes.

[0006] The core idea of ​​this disclosure is to undertake a fundamental paradigm shift: transforming the modeling object from a static "knowledge state" to a dynamic "cognitive path." It no longer focuses solely on what learners "know," but delves deeply into "how they think." This is achieved by capturing, analyzing, and modeling the entire sequence of micro-behaviors of learners' interactions with the learning interface in real time, thereby opening the "black box" of the cognitive process and enabling proactive, precise, and personalized instructional interventions.

[0007] To achieve the above objectives, the technical solution disclosed herein is summarized as follows:

[0008] On the one hand, this disclosure provides a method for modeling and predicting dynamic cognitive paths, the method comprising:

[0009] Data collection steps: Collect learners' interactive behaviors on the learning interface in real time, and generate an original interaction sequence containing multiple ordered event tuples with timestamps, wherein each event tuple contains at least an event timestamp, an event type, and an identifier of the interactive interface element.

[0010] Analysis steps: Based on the preset behavior analysis rule base, the original interaction sequence is analyzed in a streaming manner, and one or more event tuples are aggregated or transformed into a semantic micro-behavior with preset cognitive psychology meaning, thereby generating a semantic micro-behavior sequence.

[0011] Modeling steps: Input the semantic micro-behavioral sequence into a sequence modeling network to generate a dynamic cognitive path vector at each time step that encodes the complete interaction history up to the current time step;

[0012] Application steps: Based on the dynamic cognitive path vector, downstream task processing is performed through at least one prediction head to output real-time intervention signals or cognitive diagnostic reports for learners.

[0013] On the other hand, a dynamic cognitive path modeling and prediction system is provided, which includes:

[0014] The real-time interactive sequence acquisition module is used to collect learners' interactive behaviors on the learning interface in real time and generate an original interactive sequence containing multiple ordered event tuples with timestamps.

[0015] The semantic micro-behavior parsing module is used to perform streaming parsing on the original interaction sequence based on a preset behavior parsing rule base to generate a semantic micro-behavior sequence.

[0016] The dynamic cognitive path modeling module is used to input the semantic micro-behavioral sequence into a sequence modeling network and generate a dynamic cognitive path vector that encodes the complete interaction history up to the current time step at each time step.

[0017] The prediction and intervention application module is used to perform downstream task processing based on the dynamic cognitive path vector through at least one prediction head, so as to output real-time intervention signals or cognitive diagnostic reports for learners.

[0018] In another aspect, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the method described above.

[0019] In another aspect, an electronic device including a processor is also provided, the processor being configured to perform the method described above.

[0020] This disclosure fundamentally solves the problem of information scarcity in traditional methods by collecting process-oriented micro-behavioral data, providing an unprecedented high-fidelity data foundation for understanding learners' true cognitive processes. The interpretability of the cognitive process is enhanced by introducing an intermediate layer of "semantic micro-behavior," enabling the model not only to make accurate predictions but also to provide interpretable diagnoses consistent with cognitive psychology theory, answering the question "why mistakes occurred." From delayed feedback to proactive intervention, through dynamic modeling of cognitive paths, this disclosure can predict learners' impending difficulties, thereby achieving proactive and timely personalized intervention. The advanced and integrated technical architecture creatively combines knowledge graphs with Transformer sequence models, leveraging deep learning's ability to capture dynamic dependencies while incorporating structured prior knowledge of the domain, enabling forward-looking and precise personalized teaching interventions. Attached Figure Description

[0021] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0022] Figure 1 This is a flowchart of a dynamic cognitive path modeling and prediction method based on this disclosure;

[0023] Figure 2 This is a schematic diagram of a dynamic cognitive path modeling and prediction system structure corresponding to the method disclosed in this paper;

[0024] Figure 3 This is a schematic diagram of the overall system architecture and data flow of one embodiment of this disclosure;

[0025] Figure 4 This is an internal architecture diagram of a K-Transformer model according to an embodiment of this disclosure. Detailed Implementation

[0026] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0027] The technical solutions of this disclosure are illustrated below through specific embodiments. It should be understood that one or more steps mentioned in this disclosure do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of this disclosure. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not to limit the order of each method or to limit the scope of implementation of this disclosure. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of implementation of this disclosure.

[0028] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0029] To address the shortcomings of traditional technologies, this disclosure proposes a dynamic cognitive path modeling and prediction method, such as... Figure 1 As shown, the method includes:

[0030] Data collection step 101: Collect learners' interactive behaviors on the learning interface in real time and generate an original interaction sequence containing multiple ordered event tuples with timestamps, wherein each event tuple contains at least an event timestamp, an event type, and an identifier of the interactive interface element.

[0031] Analysis Step 102: Based on the preset behavior analysis rule base, perform streaming analysis on the original interaction sequence, aggregate or transform one or more event tuples into a semantic micro-behavior with preset cognitive psychology meaning, thereby generating a semantic micro-behavior sequence;

[0032] Modeling step 103: Input the semantic micro-behavioral sequence into a sequence modeling network to generate a dynamic cognitive path vector at each time step that encodes the complete interaction history up to the current time step;

[0033] Application step 104: Based on the dynamic cognitive path vector, downstream task processing is performed through at least one prediction head to output real-time intervention signals or cognitive diagnostic reports for the learner.

[0034] In the acquisition step, each event tuple also includes a knowledge point identifier associated with the identifier of the interactive interface element.

[0035] The parsing steps specifically include:

[0036] A finite state machine-based parsing engine is used to perform pattern matching and aggregation on the original interaction sequence according to the behavior parsing rule base;

[0037] The behavior parsing rule base contains rules that map physical events to semantic micro-behaviors, which include at least one of the following: reading the question stem, thinking pause, self-correction, seeking formulas, input burst, and submission attempt.

[0038] The sequence modeling network used in the modeling step is an encoder network based on the Transformer architecture.

[0039] The modeling step, before inputting the semantic micro-behavioral sequence into the encoder network, further includes:

[0040] Knowledge graph enhancement steps: For the knowledge point identifier associated with each behavior in the semantic micro-behavioral sequence, a pre-constructed subject knowledge graph is used to enhance the initial vector representation of the knowledge point through a graph attention network (GAT) to incorporate its contextual neighborhood information in the knowledge graph and generate an enhanced knowledge point vector;

[0041] The vector representation of the semantic micro-behavior is fused with the enhanced knowledge point vector and used as the input to the encoder network.

[0042] The application steps specifically include at least one of the following:

[0043] Real-time dilemma prediction: The dynamic cognitive path vector is input into a classification prediction head for predicting the next most likely semantic micro-behavior. When a negative cognitive state with a high probability is predicted, proactive intervention is triggered.

[0044] Cognitive bias diagnosis: At the end of the learning task, the final dynamic cognitive path vector is input into a multi-label classification prediction head for diagnosing cognitive biases, and the learner outputs the types of cognitive biases and their probabilities that exist in the problem-solving process.

[0045] Based on the above methods, this disclosure proposes a dynamic cognitive path modeling and prediction system, such as... Figure 2 As shown, the system includes:

[0046] The real-time interactive sequence acquisition module 201 is used to acquire learners' interactive behaviors on the learning interface in real time and generate an original interactive sequence containing multiple ordered event tuples with timestamps.

[0047] The semantic micro-behavior parsing module 202 is used to perform streaming parsing on the original interaction sequence based on a preset behavior parsing rule base to generate a semantic micro-behavior sequence;

[0048] The dynamic cognitive path modeling module 203 is used to input the semantic micro-behavioral sequence into a sequence modeling network and generate a dynamic cognitive path vector that encodes the complete interaction history up to the current time step at each time step.

[0049] The prediction and intervention application module 204 is used to perform downstream task processing based on the dynamic cognitive path vector through at least one prediction head, so as to output real-time intervention signals or cognitive diagnostic reports for learners.

[0050] The sequence modeling network in the dynamic cognitive path modeling module is a knowledge graph-enhanced Transformer network, which is configured to enhance the context of related knowledge points in the sequence using a subject knowledge graph before encoding the semantic micro-behavioral sequence.

[0051] Based on the above system architecture, such as Figure 3 As shown, the complete data flow processing procedure of the system proposed in this disclosure is as follows:

[0052] In this embodiment, the front-end learning interface includes three areas: question presentation, answer interaction, and auxiliary tools; four core processing modules: module 201: real-time interactive sequence acquisition module, module 202: semantic micro-behavioral analysis module, module 203: dynamic cognitive path modeling module, and module 204: prediction and intervention application module; knowledge and rule base: providing subject knowledge graphs, behavioral analysis rules, and intervention strategies; intervention decision engine: making intervention decisions based on prediction results and the rule base; and back-end dashboard: providing learning progress diagnostic reports to teachers and students.

[0053] Module 201: Real-time Interactive Sequence Acquisition Module; This module is the data source for the entire system and is responsible for recording all interactive behaviors of learners on the front-end interface in a non-intrusive and high-fidelity manner.

[0054] Implementation method: By embedding a lightweight JavaScript or SDK listener in the front-end learning application. This listener will capture all predefined Document Object Model (DOM) events.

[0055] Data structure: Each event is encapsulated into a structured event tuple. The acquisition frequency should be high enough (e.g., no less than 10Hz) to ensure that fast, minute operations can be captured.

[0056] Table 1: Event Tuples Data Structures

[0057] field name Data types describe Example timestamp Integer The event's UTC millisecond timestamp is used to construct a precise time series. 1663762801123 eventType Enum W3C standard event types or custom event types Mouse movement, key press, input change, video pause targetId String A unique identifier for interactive UI elements, used to associate behavior with context. The answer box for Question 1, the hint button for Question 1, and the concept video player. targetKC String The ID of the most direct knowledge point associated with this UI element is crucial; this field is of paramount importance. KC_035_Quadratic Formula data JSON Loads containing specific parameters of the event, and structural variations depending on the event type. {"x": 120, "y": 350}、{"key": "backspace key", "keycode": 8}

[0058] In the table above, mouse movement is indicated by MOUSE_MOVE; button press by KEY_PRESS.

[0059] Input changes: INPUT_CHANGE; Video pause: VIDEO_PAUSE;

[0060] The answer box for question 1 (answer_box_q1); the hint button for question 1 (hint_button_q1);

[0061] The concept video player;

[0062] KC_035_Quadratic Formula;

[0063] Key; Backspace; keyCode.

[0064] Output: A raw, timestamp-sorted sequence of event tuples .

[0065] Module 202: Semantic Micro-behavior Analysis Module; This module serves as a bridge between raw data and cognitive models, and its core task is to... Low- to mid-level, noisy physical events are transformed into semantic micro-behavioral sequences with clear cognitive psychological significance. .

[0066] Implementation: A streaming parsing engine based on finite state machines is used. This engine reads... And based on the rules in the behavior parsing rule base (see E2), the events are pattern matched and aggregated.

[0067] Core algorithm logic: Maintain an event buffer with a fixed time window (e.g., 2 seconds); the engine has multiple states, such as IDLE, READING, TYPING, and HESITATING. The arrival of a new event triggers a state transition and determines the HESITATING rule.

[0068] IF (currentState == TYPING AND time_since_last_keypress > T_hesitation_threshold) THEN transition_to(HESITATING) and emit_action(HESITATION)

[0069] Once a match is found, the engine will issue a semantic micro-action. And clear or update the relevant counters / buffers.

[0070] Table 2: Semantic Microbehavior Examples and their parsing rules

[0071] Behavior Cognitive meaning Triggering rule / algorithm example (pseudocode) PROBLEM_LOAD Starting to encounter new problems on(event.type == "PAGE_LOAD" and event.target.class == "problem_view") STEM_READ Reading and comprehension questions The mouse moves regularly within the question stem area, or hovers for a long time (>T_read). FORMULA_QUERY Retrieve or recall knowledge Click the "View Formula" button or switch to the tab containing formulas. HESITATION Thinking, getting stuck, or getting distracted In the answer area, the time during which the keyboard or mouse remains inactive exceeds the threshold T_hesitate. INPUT_BURST Develop a solution approach Inputting character sequences continuously and rapidly within a short period of time. SELF_CORRECTION Self-monitoring and error correction To delete more than N characters consecutively using the backspace key, or to select and replace a section of text. HINT_SEEKING Seeking external scaffolding Click the "Scaffolding" button ATTEMPT_SUBMIT Try submitting a solution Click the "Submit Answer" button

[0072] Output: A sequence of semantic micro-behaviors ,in Each All of them include the time of occurrence, type, and related knowledge points (targetKC).

[0073] Module 3: Dynamic Cognitive Path Modeling Module; this is the core technology of this disclosure. This module receives... Furthermore, it utilizes an innovative knowledge graph-enhanced Transformer network to encode it into a dynamic cognitive path vector capable of capturing the dynamic evolution of cognitive processes. ;like Figure 4 The flowchart shown illustrates the key components of the K-Transformer:

[0074] Input representation layer: processes the embeddings of behavioral sequences and KC sequences respectively;

[0075] Knowledge Fusion Layer: The knowledge graph information is fused with the KC embedding through a graph attention network (GAT).

[0076] Feature fusion: Concatenate the enhanced KC embedding with the behavior embedding and add positional encoding;

[0077] Transformer encoder: A feature extractor composed of multi-layer self-attention and feedforward networks;

[0078] Output: The final cognitive path vector is generated;

[0079] Based on the above data flow, the algorithm steps are explained in detail as follows: Input vectorization converts discrete semantic behaviors. and related knowledge points Through their respective embedding matrices and This is mapped to a low-dimensional dense vector:

[0080] ;in, Represents a behavior vector;

[0081] ;in, Represents a vector of knowledge points;

[0082] The and Both are embedding matrices, serving as a "dictionary" or "lookup table" for the model to convert symbolic data into mathematical vectors. Among them, It is a behavior embedding matrix, which is used to process semantic micro-behaviors (At). It maps discrete, textual behavior tags, such as "reading the question stem" and "thinking pause," into unique, low-dimensional mathematical vectors, i.e., behavior vectors; It is a knowledge point embedding matrix, which is used to process knowledge points associated with behavior. Yes. It maps knowledge point tags like "KC_SolveQuadratic (solving quadratic equations)" into a unique, low-dimensional mathematical vector, i.e., a knowledge point vector. These two matrices are the core tools for vectorizing the model's inputs. They are responsible for converting symbolic information (behaviors and knowledge points) that the model cannot directly compute into dense vectors that can be input into subsequent neural networks for computation.

[0083] The above It means: to transform a discrete, textual, semantic micro-action At into a mathematical vector that a machine can understand and compute. Where At (input) represents a semantic micro-action occurring at time t. This is not a number, but a category label, such as "reading the question stem," "thinking pause," or "seeking a formula." Neural network models cannot directly perform mathematical operations on these text labels. (Transformation tool): It is an embedding matrix; it can be understood as a "lookup table" or "dictionary". This table stores the unique vector representation corresponding to each possible semantic micro-behavior. (Output) represents the behavior vector; the operation of this formula is to use At as the key to look up this table. Within this framework, we find the corresponding "low-dimensional dense vector." This vector is the mathematical representation of At, capturing the intrinsic meaning of the action with a set of numbers so that subsequent sequence modeling networks can process it. Knowledge graph augmentation aims to help the model understand the intrinsic connections between knowledge points (e.g., "Pythagorean theorem" is prerequisite knowledge for "solving right triangles"). The method involves not directly using... Instead, it utilizes graph attention networks in subject knowledge graphs. The context is enhanced; the specific process involves the initial embedding of all knowledge points. The input node features serve as input to GAT. GAT learns the importance of each node's neighbors through an attention mechanism and aggregates neighbor information to update the representation of the central node.

[0084] Mathematical expression:

[0085]

[0086] in It is an enhanced knowledge point vector, which not only contains The information itself also incorporates information about "neighborhood relationships" from the knowledge graph.

[0087] Input fusion and positional encoding concatenate the behavior vector and the enhanced knowledge point vector:

[0088]

[0089] in This indicates a vector concatenation operation.

[0090] In order for the model to perceive the sequence information, Add a position encoding vector :

[0091]

[0092] The Transformer encoder will sequence The input is fed into a Transformer Encoder consisting of L stacked layers. At the core of each layer is a multi-head self-attention mechanism, which allows the model to simultaneously attend to information from all other positions in the sequence at each position and compute the strength of dependencies between them. This is crucial for capturing long-range dependencies (such as how initial question-reading influences final answer correction); its mathematical expression is:

[0093]

[0094] in, These are query, key, and value matrices, respectively. It is the dimension of the key vector.

[0095] After L-layer encoding, at time step The final output is the dynamic cognitive path vector. . It is a high-dimensional vector representing the entire historical interaction sequence. Rich information, context-aware numerical representation.

[0096] Output at each time step Output a dynamic cognitive path vector .

[0097] Module 204: Prediction and Intervention Application Module; this module represents the ultimate value of this disclosure. It is based on... As input, various downstream tasks can be accomplished by attaching different prediction heads.

[0098] Implementation: On top of the K-Transformer model backbone, multiple simple fully connected neural networks are connected in parallel, with each network corresponding to a task.

[0099] Table 3: Based on Downstream application tasks

[0100]

[0101] Implementation method 1:

[0102] Implementation Method Two:

[0103] Implementation method three:

[0104] Implementation Method 4:

[0105] The intervention decision engine (G) receives real-time probability output from the prediction module; based on the intervention strategy rule base (E3), it determines whether, when, and how to intervene. Example rule:

[0106] IF Predict(NextAction) == HESITATION AND P > 0.85 ANDTimeSinceLastAction > 15s THEN ShowHint(Level=1)

[0107] Its outputs include: real-time intervention signals for the front end and structural chemical condition diagnosis reports for the back end.

[0108] The following is a specific implementation scenario, using an online junior high school mathematics learning platform as an example. This platform provides online practice functions for algebra problems. We will use a student named "Wes" solving a quadratic equation on this platform as an example to explain the workflow of this disclosure in detail. Let: The learner be Wes; the learning task be solving a quadratic equation: Learning interface: The interface consists of three areas:

[0109] 1) Problem stem display area (problem_stem); 2) Answer input box (answer_box); 3) Auxiliary tools area, containing a "formula_lookup_button", clicking which can view the formula for finding the roots of a quadratic equation.

[0110] Related Knowledge Points (KC):

[0111] KC_SolveQuadratic: Solving quadratic equations in one variable (core knowledge points of the problem)

[0112] KC_QuadraticFormula: Quadratic Formula (Auxiliary Tool Related to Key Concepts)

[0113] KC_Factoring: Factoring (Learning points about potential solutions)

[0114] The implementation steps of the invention based on the technical solution disclosed herein are as follows:

[0115] Step A: Real-time Interaction Sequence Acquisition (Implementation of Module 201); When Wes enters the exercise page, the system's front-end listener starts working. Wes's series of behaviors during the problem-solving process are captured frequently. The table below shows a segment of the actual original interaction event stream. :

[0116] Table 4: Original sequence of interactive events in Wes' problem-solving process (example fragment)

[0117] timestamp eventType targetId targetKC data ...1000 PAGE_LOAD problem_view KC_SolveQuadratic {} ...1050 MOUSE_MOVE problem_stem KC_SolveQuadratic {"x":150,"y":80} ...1200 MOUSE_MOVE problem_stem KC_SolveQuadratic {"x":350,"y":85} ...2500 MOUSE_MOVE answer_box KC_SolveQuadratic {"x":200,"y":200} ...2600 MOUSE_DOWN answer_box KC_SolveQuadratic {} ...4800 KEY_PRESS answer_box KC_SolveQuadratic {"key": "x"} ...5500 KEY_PRESS answer_box KC_SolveQuadratic {"key": "Backspace"} ...9500 MOUSE_DOWN formula_lookup_button KC_QuadraticFormula {} ...15000 KEY_PRESS answer_box KC_SolveQuadratic {"key": "x"} ...15200 KEY_PRESS answer_box KC_SolveQuadratic {"key": "="} ...15400 KEY_PRESS answer_box KC_SolveQuadratic {"key": "2"} ...16000 KEY_PRESS answer_box KC_SolveQuadratic {"key": ";"} ...18000 CLICK submit_button KC_SolveQuadratic {}

[0118] Step B: Semantic micro-behavior parsing (implementation of module 202); the background semantic micro-behavior parsing module receives data in a streaming manner. Based on preset parsing rules (e.g., consecutive MOUSE_MOVE events in the problem_stem region are aggregated into STEM_READ; an interval of no operation exceeding 2 seconds is defined as HESITATION), the above original event sequence is parsed into a semantic micro-behavioral sequence with cognitive significance. :

[0119] Table 5: Semantic micro-behavioral sequences generated by parsing; where, the behavior is represented by... represent

[0120] Serial number (t) Behavior Related knowledge point $KC_t$ Analysis basis 1 PROBLEM_LOAD KC_SolveQuadratic PAGE_LOAD event 2 STEM_READ KC_SolveQuadratic It aggregates mouse movement events from ...1050 to ...1200`. 3 HESITATION KC_SolveQuadratic A silence period of more than 2 seconds between ...2600 and ...4800 4 SELF_CORRECTION KC_SolveQuadratic The input x of 4800 was deleted by the backspace of 5500. 5 FORMULA_QUERY KC_QuadraticFormula ...the CLICK event of 9550 on formula_lookup_button 6 INPUT_BURST KC_SolveQuadratic It aggregates consecutive inputs from...15000 to...16000. 7 ATTEMPT_SUBMIT KC_SolveQuadratic The CLICK event at 18000 on submit_button

[0121] Step C: Dynamic cognitive path modeling (implementation of module 203); K-Transformer model by time step Process in sequence Sequence. In Time (model processing HESITATION):

[0122] Input: Sequences (PROBLEM_LOAD, STEM_READ, HESITATION) and their associated KC sequences; Knowledge Fusion: At this point, all KCs are KC_SolveQuadratic, and the GAT layer mainly serves to consolidate the context of the current knowledge point; Self-Attention Mechanism: The attention head associates the HESITATION behavior with the previous STEM_READ behavior, and the model begins to "understand" that the student is stuck in thinking or struggling after reading the questions; Output: Cognitive path vector This vector numerically encodes a cognitive state of having read the question but getting stuck.

[0123] exist Time (Model processing FORMULA_QUERY):

[0124] Input: sequence(..., HESITATION, SELF_CORRECTION, FORMULA_QUERY).

[0125] Knowledge Fusion (Key Step): When processing the FORMULA_QUERY action, the associated knowledge point is KC_QuadraticFormula. Through the GAT layer, the model not only receives the embedding of this knowledge point but also its strong association information with KC_SolveQuadratic (KC from the previous step) in the knowledge graph (e.g., the relationship of is_method_for). The model thus understands that this formula lookup action is highly relevant to the current problem-solving task, rather than an irrelevant random action.

[0126] Self-attention mechanism: The attention mechanism calculates a very high attentional weight between the FORMULA_QUERY action and the previous HESITATION and SELF_CORRECTION actions. This allows the model to establish a key cognitive chain: "Because of stuttering and initial failure, seek help from the formula."

[0127] Output: Cognitive path vector The vector now encodes a more complex state: "the path a student takes to solve the core problem by seeking external knowledge (formulas) after failing in an initial attempt."

[0128] Step D: Prediction and Intervention Application (Implementation of Module 204); Real-time Intervention (in (occurs at any moment): When the model generates cognitive path vectors After the "stuttering" state is encoded, it is immediately sent to the "real-time dilemma prediction" header. Based on its training experience on massive amounts of historical data (i.e., many students will look up formulas or ask for hints when they stutter), this prediction header outputs a probability distribution: {P(HESITATION)=0.85, P(FORMULA_QUERY)=0.7, P(HINT_SEEKING)=0.6, ...}. The intervention decision engine receives this prediction result. According to the rule IF P(HESITATION) > 0.8 AND TimeOnTask > 10s THEN TriggerIntervention, the system triggers a proactive, non-intrusive intervention.

[0129] Specifically, a gentle prompt pops up in the corner of the learning interface: "Have you encountered any difficulties? Do you need to review the quadratic formula?" This is much more effective than waiting for Wes to completely give up or ultimately answer incorrectly before providing feedback.

[0130] Post-operative diagnostic report (in) (Generated after submission):

[0131] After Wes submits the answer, the system uses the final cognitive path vector. The data is fed into the "Cognitive Bias Diagnosis" header. The model outputs the diagnostic results and generates a learning report for Wes and his teacher, as shown in Table 6 below:

[0132] Table 6 is a learning progress report for Wes.

[0133] project content Reporting Subjects Wes's solution report for this problem Final result correct Cognitive Path Analysis - Advantages After encountering initial difficulties, they are able to proactively seek and correctly utilize the "quadratic formula" as a tool to solve the problem. Cognitive Path Analysis - Areas for Improvement There was a considerable hesitation at the beginning of the problem-solving process (lasting about 6 seconds), and the solution was only provided fluently after consulting the formula. This indicates that the user's memorization and familiarity with the quadratic formula may be insufficient. Personalized suggestions Here's a set of quick application exercises for the "quadratic formula" to help you improve your problem-solving speed and fluency.

[0134] This disclosure proposes an innovative method and system for dynamic cognitive path modeling and prediction. By establishing a semantic micro-behavioral analysis module, this disclosure achieves deep analysis from "physical events" to "cognitive behaviors," significantly improving the cognitive fidelity and signal-to-noise ratio of learning process analysis. Employing a streaming parsing engine based on Finite State Machines (FSM) and a behavioral analysis rule base, low-order, noisy raw interaction sequences (such as MOUSE_MOVE, KEY_PRESS) are aggregated and transformed into semantic micro-behaviors (such as STEM_READ, HESITATION, SELF_CORRECTION) with clear cognitive psychological meaning. This transformation process is like installing a cognitive decoder on the raw data. It filters out meaningless noise and strings together scattered operations into meaningful cognitive units, ensuring that the input for subsequent modeling is no longer cold physical signals, but a highly condensed, high signal-to-noise ratio, and high cognitive fidelity semantic behavioral sequence, laying a solid data foundation for accurately understanding the learner's true thought process. This disclosure also introduces knowledge graph-enhanced sequence modeling, enabling the cognitive path model to possess contextual awareness of subject-specific logic, significantly improving the accuracy and interpretability of model predictions. Through innovative design in the dynamic cognitive path modeling module, this disclosure incorporates a knowledge graph enhancement step before the Transformer encoder. Graph Attention Network (GAT) is used to learn the subject-specific knowledge graph, integrating neighborhood information (such as predecessors, successors, and similar relationships) of knowledge points into their vector representations to generate enhanced knowledge point vectors. This allows the model, when processing an action, not only to know which knowledge point (KC) the action is associated with, but also to understand the "position" and "relationship" of that knowledge point within the entire knowledge network. As shown in the embodiment, the model can understand that the KC_QuadraticFormula associated with the FORMULA_QUERY action is a "method" for solving the core problem KC_SolveQuadratic, rather than an irrelevant knowledge point jump. This context-aware capability enables the model to construct cognitive chains that better align with human learning logic, thereby significantly improving the accuracy of predicting learners' subsequent behaviors and judging their cognitive states, and providing stronger interpretability for the model's decisions. This disclosure utilizes a dynamic cognitive path modeling module to generate dynamic cognitive path vectors at each time step, and inputs these vectors into the "Real-time Dilemma Prediction" header of the prediction and intervention application module. The dynamic cognitive path vector encoding carries all historical information at the current moment, making it a dynamically updated "cognitive snapshot." Based on this, the system can predict the learner's most likely next behavior in real time. When a high-probability negative cognitive state (such as hesitation) is predicted, the system can trigger proactive, timely, and precise interventions (such as the gentle prompts in the embodiments). The core of the entire system is generating high-quality dynamic cognitive path vectors.Based on this general and information-rich vector representation, the system supports diverse downstream tasks by connecting multiple different prediction heads (such as classification heads, regression heads, and multi-label classification heads) in parallel. Specific types of negative cognitive states include periods of mental stagnation.

[0135] In summary, the dynamic cognitive path modeling and prediction method and system proposed in this disclosure, compared to existing technologies, is not simply a matter of functional aggregation or parameter optimization, but rather represents a fundamental technological breakthrough in three levels: data analysis, model construction, and application paradigms in learning process analysis. It enhances the fidelity of analysis through semantic micro-behavioral analysis, achieves deep modeling with subject-specific logic through a knowledge graph-enhanced Transformer network, transforms teaching intervention from passive to active through real-time prediction, and ensures the efficiency and flexibility of the solution through a modular, multi-task system architecture. These beneficial effects work together to enable this disclosure to characterize learners' cognitive paths with unprecedented accuracy and dynamics, and to provide intelligent and personalized support accordingly.

[0136] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the methods described above; the electronic device includes the processor, which is configured to perform the methods described above.

[0137] The example electronic devices described in this disclosure are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely examples and are not intended to limit the implementations of this disclosure described and / or claimed herein.

[0138] The electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for the operation of the electronic device. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0139] Multiple components in an electronic device are connected to an I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0140] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as traffic control methods. For example, in some embodiments, the traffic control method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the traffic control method described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform the traffic control method by any other suitable means (e.g., by means of firmware).

[0141] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0142] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or a traffic control device, such that when executed by the processor or controller, the program code causes the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0143] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0144] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0145] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0146] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0147] The foregoing description of specific exemplary embodiments of this disclosure is for illustrative and explanatory purposes. These descriptions are not intended to limit this disclosure to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this disclosure and their practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this disclosure, as well as various different choices and variations. The scope of this disclosure is intended to be defined by the claims and their equivalents.

Claims

1. A method for modeling and predicting dynamic cognitive paths, characterized in that, Includes the following steps: Data collection steps: Real-time collection of learners' interactive behaviors on the learning interface, generating an original interaction sequence containing multiple ordered event tuples with timestamps, wherein each event tuple contains at least an event timestamp, an event type, an interface element identifier of the interaction, and a knowledge point identifier associated with the interface element identifier of the interaction. Analysis steps: Based on the preset behavior analysis rule base, the original interaction sequence is analyzed in a streaming manner, and one or more event tuples are aggregated or transformed into a semantic micro-behavior with preset cognitive psychology meaning, thereby generating a semantic micro-behavior sequence. Modeling steps: The semantic micro-behavioral sequence is input into a knowledge graph-enhanced Transformer network, which includes: an input representation layer: processing the embeddings of the behavioral sequence and the KC sequence respectively; a knowledge fusion layer: fusing knowledge graph information with KC embeddings through a graph attention network (GAT); feature fusion: concatenating the enhanced KC embeddings with the behavioral embeddings and adding positional encoding; a Transformer encoder: a feature extractor composed of multiple layers of self-attention and feedforward networks; output: finally generating a cognitive path vector; at each time step, a dynamic cognitive path vector encoding the complete interaction history up to the current time step is generated. Application steps: Based on the dynamic cognitive path vector, downstream task processing is performed through at least one prediction head to output real-time intervention signals or cognitive diagnostic reports for learners.

2. The method according to claim 1, characterized in that, The parsing steps specifically include: A finite state machine-based parsing engine is used to perform pattern matching and aggregation on the original interaction sequence according to the behavior parsing rule base; The behavior parsing rule base contains rules that map physical events to semantic micro-behaviors, which include at least one of the following: reading the question stem, thinking pause, self-correction, seeking formulas, input burst, and submission attempt.

3. The method according to claim 1, characterized in that, The sequence modeling network used in the modeling step is an encoder network based on the Transformer architecture.

4. The method according to claim 3, characterized in that, The modeling step, before inputting the semantic micro-behavioral sequence into the encoder network, further includes: Knowledge graph enhancement steps: For the knowledge point identifier associated with each behavior in the semantic micro-behavioral sequence, a pre-constructed subject knowledge graph is used to enhance the initial vector representation of the knowledge point through a graph attention network (GAT) to incorporate its contextual neighborhood information in the knowledge graph and generate an enhanced knowledge point vector; The vector representation of the semantic micro-behavior is fused with the enhanced knowledge point vector and used as the input to the encoder network.

5. The method according to claim 1, characterized in that, The application steps specifically include at least one of the following: Real-time dilemma prediction: The dynamic cognitive path vector is input into a classification prediction head for predicting the next most likely semantic micro-behavior. When a negative cognitive state with a high probability is predicted, proactive intervention is triggered. Cognitive bias diagnosis: At the end of the learning task, the final dynamic cognitive path vector is input into a multi-label classification prediction head for diagnosing cognitive biases, and the learner outputs the types of cognitive biases and their probabilities that exist in the problem-solving process.

6. A dynamic cognitive path modeling and prediction system, characterized in that, include: The real-time interactive sequence acquisition module is used to collect learners' interactive behaviors on the learning interface in real time and generate an original interactive sequence containing multiple ordered event tuples with timestamps. The semantic micro-behavior parsing module is used to perform streaming parsing on the original interaction sequence based on a preset behavior parsing rule base to generate a semantic micro-behavior sequence. The dynamic cognitive path modeling module is used to input the semantic micro-behavioral sequence into a knowledge graph-enhanced Transformer network. This network includes: an input representation layer that processes the embeddings of the behavioral sequence and the KC sequence respectively; a knowledge fusion layer that fuses the knowledge graph information with the KC embeddings using a graph attention network (GAT); a feature fusion layer that concatenates the enhanced KC embeddings with the behavioral embeddings and adds positional encoding; a Transformer encoder that is a feature extractor composed of multiple layers of self-attention and feedforward networks; and an output that finally generates a cognitive path vector. At each time step, a dynamic cognitive path vector encoding the complete interaction history up to the current time step is generated. The prediction and intervention application module is used to perform downstream task processing based on the dynamic cognitive path vector through at least one prediction head, so as to output real-time intervention signals or cognitive diagnostic reports for learners.

7. The system according to claim 6, characterized in that, Each event tuple contains at least an event timestamp, an event type, an identifier of the interactive UI element, and a knowledge point identifier associated with the interactive UI element identifier.

8. The system according to claim 6, characterized in that, The sequence modeling network in the dynamic cognitive path modeling module is a knowledge graph-enhanced Transformer network, which is configured to enhance the context of related knowledge points in the sequence using a subject knowledge graph before encoding the semantic micro-behavioral sequence.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1 to 5.

10. An electronic device comprising a processor configured to perform the method as claimed in any one of claims 1 to 5.

Images