Standard thinking map-based all-discipline problem-solving cognitive precision diagnosis method, system and device

Abstract

The application discloses a kind of full-discipline problem solving cognitive precision diagnosis method, system and equipment based on standard thinking map, belong to artificial intelligence, educational big data mining and natural language processing technical field, method includes: loading standard thinking map with six-dimensional cognitive node as benchmark;Student original answer is parsed into cognitive action sequence by pre-training model;Based on semantic similarity, construct bipartite graph, use dynamic programming algorithm to align student action sequence with atlas optimum;In the alignment process, encounter skip step, then trigger large language model reasoning and insert implicit cognitive action to complete logical breakpoint;After alignment, traverse mapping pair, calculate weighted Euclidean distance to lock the first cognitive defect breakpoint deviating from threshold and stop subsequent tracing;Finally, retrieve and generate structured diagnostic report from six-dimensional error label library. Through implicit logic completion and first failure node locking, the deep cognitive precision attribution of non-standardized answer is realized, and the diagnostic reliability and granularity are greatly improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, educational big data mining and natural language processing, and in particular to a method, system and device for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping. Background Technology

[0002] With the development of AI education technology, automated grading has been widely applied in various assessment systems. Currently, some cutting-edge technologies can construct "thinking pathway maps" for standard answers. However, in real educational scenarios, students' answers are often non-standardized, disjointed, or contain implicit cognitive errors. How to break down disciplinary barriers and accurately anchor unstructured student answers from all subjects (such as logical derivation in mathematics, force analysis in physics, and material integration in humanities) onto standard cognitive maps, while stripping away surface-level calculation or writing errors to precisely pinpoint deep-seated "cognitive deficits," is a core technical challenge that urgently needs to be addressed in the field of educational assessment and automated diagnosis.

[0003] Existing automatic grading and diagnostic models are mostly developed in a "chimney" manner for a single subject (such as a geometric parser or grammar checker for a specific field). The underlying logic cannot be reused, and a unified cognitive standard cannot be used to diagnose question types in various subjects at the same time.

[0004] Existing technologies can typically only determine that "the final result is incorrect" or point out that "the formula in step X is incorrect," but cannot perform in-depth cognitive attribution. The system cannot accurately determine whether a student lacks a structured overall logical framework or simply has a problem with memorizing basic knowledge.

[0005] When faced with students' common mistakes such as skipping steps, using non-standard preparation methods, or having hidden logical errors, traditional string matching or single-line step comparison is prone to misjudgment, causing the marking system to fail to achieve accurate alignment of the thinking logic layer.

[0006] Therefore, there is an urgent need in this field for a general chemical condition diagnosis technology that can break down disciplinary boundaries, possess deep cognitive attribution capabilities, and robustly handle non-standardized student responses.

[0007] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide a general chemical condition diagnosis technology solution that can break down disciplinary boundaries, possess deep cognitive attribution capabilities, and robustly handle non-standardized student responses.

[0009] To achieve the above objectives, the present invention provides the following solution: A comprehensive, interdisciplinary, cognitive diagnostic method for problem-solving based on standard mind mapping includes the following steps: The system acquires a target question and loads a standard mind map corresponding to the target question. The standard mind map is a directed acyclic graph containing multiple standard cognitive nodes and directed edges between them. Each standard cognitive node corresponds to a standard cognitive action required to solve the problem, and each standard cognitive node is pre-labeled with node attributes. The node attributes include at least the cognitive dimension it belongs to and the node's feature vector within that cognitive dimension. The cognitive dimensions are uniformly divided into six dimensions: system thinking, contextual feature recognition, model pattern recognition, transformation strategy selection, cognitive operation execution, and knowledge tool invocation, forming a universal cognitive diagnostic benchmark framework. The feature vector is formed by concatenating the deep semantic encoding of the corresponding cognitive action with the one-hot encoding of the corresponding dimension, used to uniquely represent the position and attributes of the cognitive node in a multi-dimensional semantic space. The process involves acquiring students' raw answer data and preprocessing it to extract a sequence of problem-solving steps. Using a pre-trained sequence labeling model, each step in the sequence is transformed into a corresponding student cognitive action, and a sequence of student cognitive actions is constructed. Each student cognitive action is represented as a tuple containing action type and action parameters. The raw answer data includes text-based answer records or image-based answer records converted by a handwriting recognition engine. The preprocessing includes removing irrelevant characters, step segmentation, and standardization. The semantic similarity between the student's cognitive action sequence and each of the standard cognitive nodes in the standard mind graph is calculated, and a weighted bipartite graph is constructed based on the semantic similarity. An optimal matching path is then found on the bipartite graph using a dynamic programming-based graph matching algorithm to align the student's cognitive action sequence to one or more standard cognitive node sequences in the standard mind graph. The semantic similarity is obtained by calculating the cosine similarity between the vector representation of the student's cognitive action and the feature vector of the standard cognitive node. In the weighted bipartite graph, the node set is divided into a student cognitive action node set and a standard cognitive node set, and the edge weight is the semantic similarity value. Under global optimality constraints, the dynamic programming-based graph matching algorithm outputs an alignment path with the minimum total node mapping cost. During the alignment process, if the graph matching algorithm fails to find a standard cognitive node matching the current student's cognitive action in the standard mind graph under a preset node matching threshold constraint, an implicit logic completion mechanism is triggered. This implicit logic completion mechanism utilizes a pre-trained first language model to infer and generate at least one implicit cognitive action to connect the preceding and following logical steps, based on the standard cognitive nodes corresponding to the previous and following aligned actions of the current student's cognitive action. This implicit cognitive action is then inserted into the corresponding position in the standard mind graph to bridge the logical breakpoints in the graph matching algorithm. During inference, the pre-trained first language model receives text descriptions containing the preceding and following standard cognitive nodes, the question context, and prompts indicating the student's already completed steps. It then outputs an instance of an implicit cognitive action with complete action type and parameters, thereby explicitly reconstructing the omitted intermediate steps in the student's thinking process. After aligning the student's cognitive action sequence with the standard mind map, the system traverses each mapping pair on the alignment path and calculates the weighted Euclidean distance in multidimensional space between the feature vector of the student's cognitive action and the feature vector of the corresponding standard cognitive node in the mapping pair. When the weighted Euclidean distance is greater than a preset dimensional deviation threshold, the standard cognitive node in the mapping pair is marked as a failed node, and the tracing of subsequent propagating errors is stopped. The first marked failed node is determined as the cognitive defect breakpoint. After stopping the tracing, the system no longer performs deviation calculations on the alignment path after the cognitive defect breakpoint to prevent cascading interference from subsequent errors caused by the first error on the diagnostic results. The node attributes of the cognitive deficit breakpoints are extracted. Based on the cognitive dimension and feature vector in these node attributes, the corresponding error type and error label are retrieved from a pre-constructed six-dimensional cognitive error label library. A structured diagnostic report is generated, containing the error type, error label, error point description, and corresponding thought process link. The structured diagnostic report is output in JSON format, and its data fields are required to include the error dimension, error type, error label, the path position of the error node in the standard mind map, the student's specific answer performance at that node, and targeted remedial suggestions.

[0010] Optionally, after loading the standard mind map corresponding to the target question, the method further includes: Analyze the subject attributes and question type attributes of the target question; Based on the subject attributes and question type attributes, a main graph and at least one alternative graph are retrieved and loaded from a graph library; wherein, the main graph corresponds to the standard recommended solution, and the alternative graphs correspond to other non-standard effective problem-solving strategies; During the alignment process using the graph matching algorithm, the matching score between the student's cognitive action sequence and the main graph as well as all the candidate graphs is calculated simultaneously. The mind map with the highest matching score is selected as the standard mind map for actual use, so as to dynamically adapt to different problem-solving paths of students.

[0011] Optionally, the student's cognitive action is a tuple containing action type and action parameters; wherein, the action type includes: identifying contextual features, calling knowledge tools, performing cognitive operations, transforming problem representations, selecting or constructing models, and performing systematic construction; the action parameters are vectorized representations of the specific knowledge entities, formulas, or concepts operated by the corresponding action.

[0012] Optionally, a pre-trained sequence labeling model is used to transform each step in the problem-solving sequence into a corresponding student cognitive action, and a student cognitive action sequence is constructed, specifically including: Each step in the problem-solving sequence is segmented and part-of-speech tagged to obtain a text sequence; The text sequence is input into a pre-trained sequence labeling model based on the Transformer architecture, which outputs step-level student cognitive actions. The pre-trained sequence labeling model adds a linear classification layer on top of the last layer. The output dimension of the linear classification layer is equal to the sum of the number of predefined action types and the dimension of action parameters. The training samples of the pre-trained sequence labeling model are problem-solving steps for various disciplines that have been labeled by experts with cognitive action types.

[0013] Optionally, the feature vector of each standard cognitive node in the standard mind map is determined as follows: Extract the standard text description corresponding to the standard cognitive node and obtain its corresponding cognitive dimension; The standard text description is semantically encoded using a pre-trained second-largest language model to obtain an initial semantic vector; at the same time, a one-hot encoding is generated for each standard cognitive node under the corresponding cognitive dimension. The initial semantic vector is concatenated with the one-hot encoding of the dimension to obtain the multidimensional feature vector of the standard cognitive node.

[0014] Optionally, the weighted Euclidean distance in multidimensional space between the feature vector of the student's cognitive action and the feature vector of the corresponding standard cognitive node in the calculation mapping pair is specifically calculated according to the following formula:

[0015] in, D For weighted Euclidean distance; NThe total dimension of the feature vector; x i For the student's cognitive action feature vector in the th i Values ​​in the dimension; y i For the standard cognitive node feature vector at the th i Values ​​in the dimension; α i For the first i The weight coefficients corresponding to each dimension are pre-set based on the causal contribution of the cognitive dimension to the problem-solving result; the weight coefficients of different cognitive dimensions satisfy the following... .

[0016] Optionally, the six-dimensional cognitive error label library specifically includes a tree-structured error label system corresponding to six cognitive dimensions: system thinking, contextual feature recognition, model pattern recognition, transformation strategy selection, cognitive operation execution, and knowledge tool invocation. Each label node under each cognitive dimension is configured with a multi-dimensional feature center vector, which is used to calculate the similarity with the feature vector of the cognitive defect breakpoint to determine the specific error label. Each non-leaf node in the tree-structured error label system represents an error category, and each leaf node represents a specific error label. The feature center vector is the mean vector of the feature vectors of all historical cognitive defect breakpoint samples belonging to the same leaf label.

[0017] A comprehensive, multidisciplinary problem-solving cognitive precision diagnostic system based on standard mind mapping includes: The graph loading module is used to obtain the target question and load the standard mind graph corresponding to the target question; wherein, the standard mind graph is a directed acyclic graph, which contains multiple standard cognitive nodes and directed edges between nodes; each standard cognitive node corresponds to a standard cognitive action required to solve the problem, and each standard cognitive node is pre-labeled with node attributes; the node attributes include at least: the cognitive dimension to which it belongs, and the feature vector of the node under the corresponding cognitive dimension; The answer parsing module is used to acquire students' original answer data and preprocess the original answer data to extract the problem-solving step sequence; using a pre-trained sequence labeling model, each step in the problem-solving step sequence is converted into a corresponding student cognitive action, and a student cognitive action sequence is constructed; wherein, each student cognitive action is a tuple containing action type and action parameters. The dynamic alignment module is used to calculate the semantic similarity between the student's cognitive action sequence and each of the standard cognitive nodes in the standard mind graph, and to construct a weighted bipartite graph based on the semantic similarity; and to solve for the optimal matching path on the bipartite graph using a graph matching algorithm based on dynamic programming, so as to align the student's cognitive action sequence to one or more standard cognitive node sequences on the standard mind graph. The logic completion module is used to trigger an implicit logic completion mechanism during the alignment process if, under a preset node matching threshold constraint, the graph matching algorithm cannot find a standard cognitive node in the standard mind graph that matches the current student's cognitive action. The implicit logic completion mechanism uses a pre-trained first language model to infer and generate at least one implicit cognitive action to connect the preceding and following logic based on the standard cognitive nodes corresponding to the previous and following aligned actions of the current student's cognitive action. The implicit cognitive action is then inserted into the corresponding position in the standard mind graph to bridge the logical breakpoints in the graph matching algorithm. The breakpoint localization module is used to traverse each mapping pair on the alignment path after aligning the student's cognitive action sequence with the standard mind map, calculate the weighted Euclidean distance in multidimensional space between the feature vector of the student's cognitive action in the mapping pair and the feature vector of the corresponding standard cognitive node; when the weighted Euclidean distance is greater than a preset dimensional deviation threshold, the standard cognitive node in the mapping pair is marked as a failed node, and the tracing of subsequent transitive errors is stopped, and the first marked failed node is determined as the cognitive defect breakpoint; The diagnosis generation module is used to extract the node attributes of the cognitive defect breakpoints, and based on the cognitive dimension and feature vector in the node attributes, retrieve the corresponding error type and error label from the pre-built six-dimensional cognitive error label library, and generate a structured diagnostic report containing the error type, error label, error point description and corresponding thought process link.

[0018] Optionally, the graph loading module further includes: parsing the subject attributes and question type attributes of the target question; and retrieving and loading a main graph and at least one alternative graph from a graph library based on the subject attributes and question type attributes; wherein the main graph corresponds to the standard recommended solution, and the alternative graphs correspond to other non-standard effective problem-solving strategies; The dynamic alignment module is also used to synchronously calculate the matching score between the student's cognitive action sequence and the main mind map and all the candidate mind maps; and select the mind map with the highest matching score as the standard mind map for actual use, so as to dynamically adapt to the different problem-solving paths of students.

[0019] A comprehensive problem-solving cognitive precision diagnostic device based on standard mind mapping, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0020] A computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0021] Compared with the prior art, the present invention has the following beneficial effects: First, it achieves a unified underlying approach to cognitive diagnosis across all disciplines. This invention constructs a six-dimensional cognitive action classification system and a standard mind map, mapping the diverse problem-solving behaviors across disciplines to six universal cognitive dimensions: systemic thinking, contextual feature recognition, model pattern recognition, transformation strategy selection, cognitive operation execution, and knowledge tool retrieval. This fundamentally breaks down the barriers of traditional siloed development of automated grading systems by discipline. A single diagnostic engine can simultaneously handle multiple subject types, including mathematics, physics, chemistry, and humanities essays, significantly reducing the development boundaries and deployment costs of educational AI assessment systems and enabling the large-scale reuse of cross-disciplinary diagnostic capabilities.

[0022] Secondly, this invention achieves a leap in diagnostic accuracy from superficial judgment to deep cognitive attribution. It introduces a multi-dimensional weighted Euclidean distance metric and a first-failure-node locking mechanism, enabling it to penetrate the chain of calculation errors on the surface and, like a domain expert, precisely trace back to the specific dimension and stage where the initial cognitive breakdown occurred. The diagnostic results no longer stop at "wrong answer" or "wrong formula step," but can clearly define whether the student's "transformation strategy selection" is inappropriate, "model pattern recognition" is flawed, or "knowledge tool retrieval" is omitted—all deep-seated cognitive deficiencies—providing a precise focus for personalized remedial teaching.

[0023] Third, it significantly improves the diagnostic robustness and reliability for non-standardized responses. This invention innovatively proposes an implicit logic completion mechanism, which intelligently reconstructs skipped steps and omitted thought processes in student responses using a pre-trained language model, completing incomplete thought flows to a level that aligns with standard graphs. Simultaneously, combined with a dynamic matching strategy for alternative graphs, it can accommodate students' non-standard problem-solving paths. The synergy of these two mechanisms effectively solves the problems of high misjudgment rates and high alignment failure rates in traditional string matching or single-step comparison when faced with skipped steps and divergent responses, making the dynamic graph alignment technology truly reliable and engineering-ready in real-world, complex educational scenarios.

[0024] Fourth, the diagnostic results are highly structured and can be directly used as the data foundation for downstream intelligent education systems. The invention outputs a JSON structure containing error dimensions, error labels, graph path locations, and targeted remedial suggestions. This structure boasts strong parsing and integrability, enabling seamless integration with learning analysis systems, adaptive learning engines, and teacher decision support dashboards, providing a high-quality, standardized diagnostic data interface for the intelligent education ecosystem. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the process for a comprehensive problem-solving cognitive accuracy diagnosis method based on standard mind mapping, provided in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the structure of a comprehensive problem-solving cognitive precision diagnosis system based on standard mind mapping, provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The purpose of this invention is to provide a general chemical condition diagnosis technology solution that can break down disciplinary boundaries, possess deep cognitive attribution capabilities, and robustly handle non-standardized student responses.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1: Reference Figure 1 , Figure 1 This invention provides a flowchart illustrating a comprehensive, multidisciplinary problem-solving cognitive accuracy diagnostic method based on standard mind mapping. The method specifically includes the following steps: Step S100: Obtain the target question and load the corresponding standard mind map.

[0032] This step is used to establish a "gold standard" for diagnosis. The system first receives the target question to be diagnosed, which can be a math geometry problem, a physics mechanics problem, or a language reading comprehension question. The system analyzes the subject attribute (e.g., mathematics) and question type attribute (e.g., problem-solving question) of the question, and then retrieves and loads the standard mind map that uniquely corresponds to the question from a pre-set mind map library.

[0033] This standard mind map is a directed acyclic graph, constructed through collaboration between subject matter experts and cognitive experts to break down the derivation process of the standard answer according to cognitive dimensions. For example, the standard solution to a mathematical proof problem can be broken down as follows: Node 1 (Systems Thinking): Clearly define the overall logical structure of the proof (e.g., "using proof by contradiction").

[0034] Node 2 (Model Pattern Recognition): Identifies the geometric model involved in the problem (such as "congruent triangle model").

[0035] Node 3 (Knowledge Tool Call): Calls specific definitions (such as "theorem for congruent triangles").

[0036] Node 4 (Context Feature Recognition): Extract the known conditions and the conclusion to be proved as features.

[0037] Node 5 (Transformation Strategy Selection): Determines how to draw auxiliary lines to construct congruent triangles.

[0038] Node 6 (Cognitive Operation Execution): Perform algebraic derivation calculations of angles and side lengths step by step.

[0039] These nodes are connected by directed edges according to the order of derivation, forming a graph.

[0040] Each node in the graph is pre-labeled with node attributes, which include at least: (1) the cognitive dimension to which it belongs; and (2) the multidimensional feature vector of the node. The feature vector is determined as follows: First, the text description of the node (such as "using the side-angle-side theorem to prove triangle congruence") is semantically encoded using a pre-trained second-largest language model to obtain an initial semantic vector; second, the cognitive dimension to which the node belongs (such as "model pattern recognition") is obtained, and a one-hot encoding for the dimension is generated; finally, the initial semantic vector and the one-hot encoding are concatenated to form the final multidimensional feature vector of the node, which contains both semantic and cognitive dimension information.

[0041] Step S200: Obtain the student's original answer data and parse it into a sequence of student cognitive actions.

[0042] This step is used to transform non-standardized student input into a standardized cognitive flow that the system can compute.

[0043] The system can obtain students' original answer data through API interfaces or direct uploads. This data can be in text format or image answer records converted by a handwriting recognition engine (such as an OCR module). Subsequently, the system preprocesses the original data, including removing irrelevant characters, segmenting steps according to common step identifiers (such as "Solution:", "1.", newline characters, etc.), and standardizing the data (such as unifying full-width and half-width characters).

[0044] Next, the system uses a pre-trained sequence labeling model to transform each segmented problem-solving step into a standard student cognitive action. This model is based on the Transformer architecture and adds a linear classification layer at the last layer. Its output dimension is equal to the sum of the number of the six predefined action types and the dimension of the action parameters.

[0045] For example, given a student's answer step "because angle A equals angle B", the model might output a tuple: (Action type: context feature recognition, action parameters: vector [angle A, angle B, equality relationship]).

[0046] For the step "So the triangle is an isosceles triangle", the output is: (Action type: Model pattern recognition, Action parameter: Vector [Isosceles triangle, Decision condition]).

[0047] For the step "From the Law of Cosines, we get...", the output is: (Action type: Knowledge tool call, Action parameter: Vector [Law of Cosines formula, Parameter mapping]).

[0048] Similarly, all of the student's consecutive answering steps are transformed into an ordered sequence of cognitive actions. The training samples for this model consist of problem-solving step texts covering hundreds of questions across all subjects, with cognitive action types and parameters manually annotated by experts.

[0049] Step S300: Calculate semantic similarity and construct a bipartite graph, and use a graph matching algorithm based on dynamic programming to achieve dynamic alignment.

[0050] This is a crucial step in achieving "high-reliability alignment of non-standard data".

[0051] The system iterates through each action in the student's cognitive action sequence and compares it with each standard cognitive node in the standard mind map, calculating the semantic similarity between them. This similarity is obtained by calculating the cosine similarity of the feature vectors of the two actions in multidimensional space. The higher the semantic similarity, the more consistent the student's cognitive behavior at that step is with that cognitive step in the standard solution.

[0052] The system treats the student's cognitive action sequence as one set of nodes and the nodes of the standard mind map as another set of nodes. It then constructs a weighted complete bipartite graph based on the semantic similarity values ​​between all node pairs. The weights of the edges in the bipartite graph represent the semantic similarity of the corresponding node pairs.

[0053] To find a globally optimal alignment path while allowing students to engage in non-standardized behaviors such as insertion, deletion, and reordering, this invention employs a graph matching algorithm based on dynamic programming. The algorithm aims to minimize the sum of the cumulative costs of node mappings (higher costs for lower semantic similarity) and solves this problem under a "global optimality constraint." Ultimately, the algorithm outputs an optimal matching path that explicitly identifies which cognitive actions in the student's sequence correspond to which standard cognitive nodes in the standard graph, and which standard cognitive steps might have been missed by the student.

[0054] Step S400: During the alignment process, trigger the implicit logic completion mechanism to bridge the jump step.

[0055] This is the core innovation that effectively solves the problem of "students skipping steps, causing misalignment".

[0056] During the execution of the graph matching algorithm, if the algorithm cannot find a standard cognitive node that matches a student's cognitive action "subsequent action" in the entire standard mind graph under the current preset high node matching threshold constraint, the system determines that the student has skipped a step or omitted thinking at this point, and immediately triggers the implicit logic completion mechanism.

[0057] This mechanism automatically invokes a pre-trained primary language model. The model receives a carefully crafted prompt containing: (1) a text description of the standard cognitive node corresponding to the previous aligned action of the student's current cognitive action; (2) a text description of the standard cognitive node corresponding to the next aligned action of the student's current cognitive action; (3) the complete context of the question; and (4) the complete sequence of steps the student has already taken to answer the question.

[0058] Guided by the aforementioned prompts, the first pre-trained large language model will perform logical reasoning and generate one or more "implicit cognitive actions" to connect the preceding and following logic. For example, if a student omits the step "according to the theorem of equal angles opposite equal sides" between "angle A = angle B" and "the triangle is an isosceles triangle", the model will generate the missing cognitive action and output it as an instance containing the complete action type and action parameters.

[0059] The system then inserts the generated implicit cognitive action into the corresponding position between two standard cognitive nodes in the standard mind graph, thus bridging the logical breakpoints in the graph. Afterward, the graph matching algorithm continues alignment on the completed standard mind graph until the entire sequence is finished.

[0060] Step S500: After alignment is completed, calculate the weighted Euclidean distance and lock the first cognitive deficit breakpoint.

[0061] After completing the alignment of the student's cognitive action sequence with the standard mind map, the system will perform a deviation diagnosis on each pair of mappings on the alignment path (i.e., a student's cognitive action and its aligned standard cognitive node).

[0062] Specifically, the system extracts the feature vectors of two nodes in the mapping pair and calculates their weighted Euclidean distance D in the multidimensional space according to the following formula:

[0063] in, N The total dimension of the feature vector ；x i For the student node feature vector at the th i Values ​​in the dimension; y i For the standard node feature vector at the th i Values ​​in the dimension; α i For the first i The weight coefficients corresponding to each dimension. These weight coefficients are not equal, but are pre-set based on the causal contribution of the cognitive dimension to which that feature dimension belongs to the final problem-solving result. For example, in mathematical calculation problems, the weight of the "cognitive operation execution" dimension is usually higher than that of the "contextual feature recognition" dimension.

[0064] When the weighted Euclidean distance of a certain pair of mappings D When the deviation from the threshold is greater than the dimensional threshold dynamically preset by the system based on the subject and question type, the system marks the standard cognitive node in the mapping pair as a "failed node".

[0065] A key innovation of this invention lies in the fact that once the first marked failure node appears, the system immediately stops calculating the deviation of subsequent mapping pairs on that alignment path after marking it as a "cognitive defect breakpoint." This is because a series of errors made by students after an early cognitive breakpoint are highly likely to be "transmissive errors" triggered by the propagation of that early error. If subsequent errors are diagnosed continuously, the "root cause error" will be submerged in a "chain reaction." By locking the first failure node and stopping the tracing of its source, this invention achieves precise capture of the original cause of the error, effectively preventing cascading interference.

[0066] Step S600: Generate and output the results of in-depth error analysis and structured diagnosis.

[0067] Finally, the system extracts the node attributes of "cognitive deficit breakpoints," especially their corresponding cognitive dimension (e.g., "knowledge tool access") and its feature vector. The system then uses this feature vector as the query object to search within a pre-built six-dimensional cognitive error tag library.

[0068] This error tag library is a six-dimensional tree-structured error tag system. If an invalid node belongs to the "Knowledge Tool Invocation" dimension, the system will search within the tree tags of that dimension. In this tree structure, non-leaf nodes represent error categories (such as "Formula memorization error"), and leaf nodes represent specific error tags (such as "Confusion of geometric series summation formula"). Each leaf tag node is configured with a multi-dimensional feature center vector, which is the mean vector of the feature vectors of all student breakpoints historically diagnosed with the same specific error.

[0069] The system finds the specific error label that best matches the feature vector of the cognitive deficit breakpoint by calculating the cosine similarity between the feature vector of each leaf label and the feature center vector of each leaf label. Finally, the system generates a structured diagnostic report.

[0070] Example 2: Reference Figure 2 , Figure 2 This is a schematic diagram illustrating the structure of a comprehensive problem-solving cognitive precision diagnosis system based on standard mind mapping, provided in an embodiment of the present invention. This system completely corresponds to the method embodiment described above and includes: Atlas loading module 101: responsible for executing step S100 and its preferred scheme, not only loading the main atlas, but also dynamically loading and selecting alternative atlases according to subject and question type attributes.

[0071] Answer Analysis Module 102: Responsible for executing step S200, which transforms the multimodal raw answers into a structured and standardized sequence of student cognitive actions.

[0072] Dynamic Alignment Module 103: Responsible for executing step S300, which achieves optimal matching between non-standard responses and standard graphs by constructing a weighted bipartite graph and executing a dynamic programming algorithm.

[0073] Logic completion module 104: Responsible for executing step S400. When an alignment breakpoint is detected, it drives the first pre-trained large language model to generate implicit cognitive actions to achieve thought flow completion.

[0074] Breakpoint location module 105: responsible for executing step S500. After alignment, it calculates the weighted Euclidean distance of all mapping pairs and executes the precise location strategy of "locking the first failed node - stopping tracing".

[0075] Diagnosis generation module 106: responsible for executing step S600, extracting breakpoint attributes and searching and matching them in the six-dimensional tree-like label library, and finally generating and outputting a structured JSON diagnostic report.

[0076] The collaborative workflow between modules is consistent with the above-described method steps and will not be repeated here. Through the above system, this invention successfully constructs a multidisciplinary cognitive diagnostic engine with high versatility, diagnostic accuracy, and robustness to non-standard data.

[0077] Example 1: Diagnosis of non-standard paths in math problems: Problem: Solve for a geometric sequence { a n} before n The sum of terms, where a1=1, q=2, n=5.

[0078] Student's answer process: 1. "Use the summation formula to calculate." (Cognitive-motor: Systems thinking and strategy selection).

[0079] 2. S 5 = 1 + 2 + 4 + 8 + 16 (Cognitive action: cognitive operation execution - the most primitive accumulation method is used here, rather than the standard formula method). 3. S 5=5 (Cognitive action: cognitive operation execution - obviously added incorrectly). 4. "The answer is 5" (cognitive action: conclusion output).

[0080] The diagnostic process of this invention: Graph Loading: The system loads the main graph for the question (standard path: scenario identification → call summation formula → substitute parameters → calculation result) and an alternative graph (cumulative path: scenario identification → item-by-item accumulation → summation result).

[0081] Dynamic Alignment: The dynamic alignment module 103 synchronously calculates the matching degree between the student's answer sequence and the main and alternative graphs. It finds that the semantic similarity between the student's second step, "listing and adding up the items," and the "item-by-item accumulation" node in the alternative graph is extremely high. The system automatically selects the alternative graph as the alignment standard.

[0082] Breakpoint localization: After alignment, the breakpoint localization module 105 traverses the mapping pairs. For the action "1+2+4+8+16", the standard result is 31. The system calculates the weighted Euclidean distance of this mapping pair, and the result far exceeds the preset threshold.

[0083] Diagnostic output: The system identifies this node as the first failed node and stops subsequent tracing. Diagnostic generation module 106 outputs a report: Analysis: This invention did not judge the student as wrong because the student used the non-standard strategy of "accumulation method". Instead, it correctly aligned the alternative graphs and accurately located the error to the specific calculation operation error, rather than the method itself, which shows a high level of diagnostic wisdom.

[0084] Example 2: In-depth attribution diagnosis of physics problems: Problem: An object slides down a smooth inclined plane. Find its acceleration.

[0085] Student's answer process: 1. The force analysis diagram only shows gravity. (Cognitive-motor: Contextual feature recognition - missing supporting forces) 2. Incorrect calculation of the resultant force leads to an incorrect final acceleration.

[0086] The diagnostic process of this invention: Logical alignment: The system first aligns the student's step 1 to the "Contextual Feature Recognition: Force Analysis" node in the standard graph.

[0087] Breakpoint location: When calculating the weighted Euclidean distance of this mapping pair, because the key entity of "slope support force" is missing in the feature vector, the distance value is huge, and this node is immediately marked as the first failure node. Subsequent steps such as "calculating the resultant force" will no longer perform deviation calculations.

[0088] Diagnostic Output: The system accurately outputs the following diagnosis: Error Dimension: "Context Feature Recognition"; Error Label: "Force Analysis - Missing Object". It did not misjudge the student as using the wrong formula or making a calculation error. This precisely verifies that the present invention completely solves the problem of "coarse attribution granularity" in the background technology.

[0089] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0090] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for accurate diagnosis of problem-solving cognition across all disciplines based on standard mind mapping, characterized in that: Includes the following steps: Obtain the target question and load the standard mind map corresponding to the target question; wherein, the standard mind map is a directed acyclic graph, which contains multiple standard cognitive nodes and directed edges between nodes; each standard cognitive node corresponds to a standard cognitive action required to solve the problem, and each standard cognitive node is pre-labeled with node attributes; the node attributes include at least: the cognitive dimension to which it belongs, and the feature vector of the node under the corresponding cognitive dimension; The original answer data of students is obtained and preprocessed to extract the problem-solving step sequence; using a pre-trained sequence labeling model, each step in the problem-solving step sequence is converted into a corresponding student cognitive action, and a student cognitive action sequence is constructed; wherein each student cognitive action is a tuple containing action type and action parameters. Calculate the semantic similarity between the student's cognitive action sequence and each of the standard cognitive nodes in the standard mind graph, and construct a weighted bipartite graph based on the semantic similarity; use a graph matching algorithm based on dynamic programming to solve for the optimal matching path on the bipartite graph to align the student's cognitive action sequence to one or more standard cognitive node sequences on the standard mind graph. During the alignment process, if the graph matching algorithm fails to find a standard cognitive node matching the current student's cognitive action in the standard mind graph under the preset node matching threshold constraint, an implicit logic completion mechanism is triggered. The implicit logic completion mechanism uses a pre-trained first language model to infer and generate at least one implicit cognitive action to connect the preceding and following logic based on the standard cognitive nodes corresponding to the previous and following aligned actions of the current student's cognitive action. The implicit cognitive action is then inserted into the corresponding position in the standard mind graph to bridge the logical breakpoints in the graph matching algorithm. After aligning the student's cognitive action sequence with the standard mind map, each mapping pair on the alignment path is traversed, and the weighted Euclidean distance between the feature vector of the student's cognitive action in the mapping pair and the feature vector of the corresponding standard cognitive node in the multidimensional space is calculated. When the weighted Euclidean distance is greater than a preset dimensional deviation threshold, the standard cognitive node in the mapping pair is marked as a failed node, and the tracing of subsequent transitive errors is stopped. The first marked failed node is determined as the cognitive defect breakpoint. Extract the node attributes of the cognitive defect breakpoints. Based on the cognitive dimension and feature vector in the node attributes, retrieve the corresponding error type and error label from the pre-built six-dimensional cognitive error label library, and generate a structured diagnostic report containing the error type, error label, error point description, and corresponding thought process link.

2. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 1, characterized in that, After loading the standard mind map corresponding to the target question, the following is also included: Analyze the subject and question type attributes of the target question; Based on the subject attributes and question type attributes, a main graph and at least one alternative graph are retrieved and loaded from a graph library; wherein, the main graph corresponds to the standard recommended solution, and the alternative graphs correspond to other non-standard effective problem-solving strategies; During the alignment process using the graph matching algorithm, the matching score between the student's cognitive action sequence and the main graph as well as all the candidate graphs is calculated simultaneously. The mind map with the highest matching score is selected as the standard mind map for actual use, so as to dynamically adapt to different problem-solving paths of students.

3. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 1, characterized in that, The student's cognitive action is a tuple containing action type and action parameters; wherein, the action type includes: identifying contextual features, calling knowledge tools, performing cognitive operations, transforming problem representations, selecting or constructing models, and performing systematic construction; the action parameters are vectorized representations of the specific knowledge entities, formulas, or concepts operated by the corresponding action.

4. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 1, characterized in that, Using a pre-trained sequence labeling model, each step in the problem-solving sequence is transformed into a corresponding student cognitive action, and a student cognitive action sequence is constructed, specifically including: Each step in the problem-solving sequence is segmented and part-of-speech tagged to obtain a text sequence; The text sequence is input into a pre-trained sequence labeling model based on the Transformer architecture, which outputs step-level student cognitive actions. The pre-trained sequence labeling model adds a linear classification layer on top of the last layer. The output dimension of the linear classification layer is equal to the sum of the number of predefined action types and the dimension of action parameters. The training samples of the pre-trained sequence labeling model are problem-solving steps for various disciplines that have been labeled by experts with cognitive action types.

5. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 1, characterized in that, The feature vectors of each standard cognitive node in the standard mind map are determined as follows: Extract the standard text description corresponding to the standard cognitive node and obtain its corresponding cognitive dimension; The standard text description is semantically encoded using a pre-trained second-largest language model to obtain an initial semantic vector; at the same time, a one-hot encoding is generated for each standard cognitive node under the corresponding cognitive dimension. The initial semantic vector is concatenated with the one-hot encoding of the dimension to obtain the multidimensional feature vector of the standard cognitive node.

6. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 1, characterized in that, The weighted Euclidean distance between the feature vectors of the student's cognitive actions and the feature vectors of the corresponding standard cognitive nodes in the multidimensional space of the computational mapping pair is specifically calculated according to the following formula: in, D For weighted Euclidean distance; N The total dimension of the feature vector; x i For the student's cognitive action feature vector in the th i Values ​​in the dimension; y i For the standard cognitive node feature vector at the th i Values ​​in the dimension; α i For the first i The weight coefficients corresponding to each dimension are pre-set based on the causal contribution of the cognitive dimension to the problem-solving result; the weight coefficients of different cognitive dimensions satisfy the following... .

7. The method for accurate diagnosis of problem-solving cognition across all subjects based on standard mind mapping as described in claim 6, characterized in that, The six-dimensional cognitive error label library specifically includes a tree-structured error label system corresponding to six cognitive dimensions: system thinking, contextual feature recognition, model pattern recognition, transformation strategy selection, cognitive operation execution, and knowledge tool invocation. Each label node under a cognitive dimension is configured with a multi-dimensional feature center vector, which is used to calculate the similarity with the feature vector of the cognitive defect breakpoint to determine the specific error label.

8. A comprehensive problem-solving cognitive precision diagnosis system based on standard mind mapping, characterized in that: include: The graph loading module is used to obtain the target question and load the standard mind graph corresponding to the target question; wherein, the standard mind graph is a directed acyclic graph, which contains multiple standard cognitive nodes and directed edges between nodes; each standard cognitive node corresponds to a standard cognitive action required to solve the problem, and each standard cognitive node is pre-labeled with node attributes; the node attributes include at least: the cognitive dimension to which it belongs, and the feature vector of the node under the corresponding cognitive dimension; The answer parsing module is used to acquire students' original answer data and preprocess the original answer data to extract the problem-solving step sequence; using a pre-trained sequence labeling model, each step in the problem-solving step sequence is converted into a corresponding student cognitive action, and a student cognitive action sequence is constructed; wherein, each student cognitive action is a tuple containing action type and action parameters. The dynamic alignment module is used to calculate the semantic similarity between the student's cognitive action sequence and each of the standard cognitive nodes in the standard mind graph, and to construct a weighted bipartite graph based on the semantic similarity; and to solve for the optimal matching path on the bipartite graph using a graph matching algorithm based on dynamic programming, so as to align the student's cognitive action sequence to one or more standard cognitive node sequences on the standard mind graph. The logic completion module is used to trigger an implicit logic completion mechanism during the alignment process if, under a preset node matching threshold constraint, the graph matching algorithm cannot find a standard cognitive node in the standard mind graph that matches the current student's cognitive action. The implicit logic completion mechanism uses a pre-trained first language model to infer and generate at least one implicit cognitive action to connect the preceding and following logic based on the standard cognitive nodes corresponding to the previous and following aligned actions of the current student's cognitive action. The implicit cognitive action is then inserted into the corresponding position in the standard mind graph to bridge the logical breakpoints in the graph matching algorithm. The breakpoint localization module is used to traverse each mapping pair on the alignment path after aligning the student's cognitive action sequence with the standard mind map, calculate the weighted Euclidean distance in multidimensional space between the feature vector of the student's cognitive action in the mapping pair and the feature vector of the corresponding standard cognitive node; when the weighted Euclidean distance is greater than a preset dimensional deviation threshold, the standard cognitive node in the mapping pair is marked as a failed node, and the tracing of subsequent transitive errors is stopped, and the first marked failed node is determined as the cognitive defect breakpoint; The diagnosis generation module is used to extract the node attributes of the cognitive defect breakpoints, and based on the cognitive dimension and feature vector in the node attributes, retrieve the corresponding error type and error label from the pre-built six-dimensional cognitive error label library, and generate a structured diagnostic report containing the error type, error label, error point description and corresponding thought process link.

9. The comprehensive problem-solving cognitive precision diagnosis system based on standard mind mapping according to claim 8, characterized in that, The graph loading module further includes: parsing the subject attributes and question type attributes of the target question; and retrieving and loading a main graph and at least one alternative graph from a graph library based on the subject attributes and question type attributes; wherein the main graph corresponds to the standard recommended solution, and the alternative graphs correspond to other non-standard effective problem-solving strategies; The dynamic alignment module is also used to synchronously calculate the matching score between the student's cognitive action sequence and the main mind map and all the candidate mind maps; and select the mind map with the highest matching score as the standard mind map for actual use, so as to dynamically adapt to the different problem-solving paths of students.

10. A comprehensive problem-solving cognitive precision diagnostic device based on standard mind mapping, characterized in that, include: A memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the method as claimed in any one of claims 1 to 7.

Images