Artificial intelligence multi-scene artificial intelligence education set teaching system

Abstract

This invention discloses an AI multi-scenario artificial intelligence education kit teaching system, belonging to the field of artificial intelligence education technology. The system includes: a multimodal data acquisition module, a cross-modal semantic understanding module, a personal knowledge graph construction module, a knowledge graph application module, and a teaching interaction module. This invention comprehensively collects learning data such as handwriting, voice, and code through multimodal data acquisition and cross-modal semantic fusion technology, breaking down information barriers. It constructs a personal knowledge graph centered on knowledge points and updates it incrementally, dynamically depicting the student's knowledge status. The system automatically detects knowledge gaps and generates personalized review paths that conform to the dependencies between knowledge points, achieving individualized instruction. Simultaneously, it provides a multi-terminal visual interactive interface to intuitively display the knowledge graph and learning records, significantly improving the effectiveness of teaching interaction and learning efficiency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence education technology, specifically to an AI multi-scenario artificial intelligence education kit teaching system. Background Technology

[0002] With the deep integration of artificial intelligence (AI) technology and education, AI education kits have become an important vehicle for schools to conduct AI teaching. Currently, AI education is transitioning from standardized to personalized teaching, generating diverse learning data during the learning process, including handwriting, voice recordings, code, operation videos, and experimental data. How to comprehensively collect and deeply mine the value of this data, and accurately depict students' knowledge mastery, has become a core issue in improving the effectiveness of AI education. However, most existing AI education systems remain at the basic level of content delivery and homework grading, failing to meet the needs of personalized and scenario-based teaching.

[0003] Existing AI-powered education and teaching systems suffer from several shortcomings: Firstly, their data collection dimensions are limited, mostly supporting only a few modalities such as text and video, failing to simultaneously collect crucial information like handwritten notes, hardware operation videos, and experimental sensor data, resulting in an incomplete portrayal of the student's learning process. Secondly, their cross-modal semantic understanding capabilities are insufficient, with different modalities of data isolated from each other, lacking a unified semantic representation method, and unable to effectively integrate multi-source data for comprehensive analysis. Thirdly, their personal knowledge graph construction methods are static, unable to dynamically update the level of knowledge mastery based on students' real-time learning behavior, and unable to accurately identify erroneous perceptions and knowledge gaps. Fourthly, their personalized recommendation mechanisms are rudimentary, failing to consider the dependencies and causal relationships between knowledge points in review path generation, and their teaching interaction methods are limited, with poor visualization effects, making it difficult to intuitively display the full picture of a student's knowledge system.

[0004] In summary, existing technologies cannot achieve comprehensive collection and deep integration of students' multimodal learning data, making it difficult to construct dynamic and accurate personal knowledge graphs and provide truly personalized teaching services, severely limiting the teaching effectiveness of AI education kits. Therefore, there is an urgent need to develop an AI multi-scenario artificial intelligence education kit teaching system that can integrate multimodal data, possess cross-modal semantic understanding capabilities, dynamically construct personal knowledge graphs, and generate personalized teaching plans. This would address the technical limitations of existing technologies and promote the development of AI education towards greater precision, efficiency, and personalization. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI multi-scenario artificial intelligence education kit teaching system. It can connect to hardware terminals through a multimodal data acquisition module to collect five types of data: handwriting, voice, code, video, and experiments. A cross-modal semantic understanding module fuses multimodal features into a unified semantic vector. A personal knowledge graph construction module uses knowledge points as the core, including equilateral attributes of mastery level, and adopts incremental updates. A knowledge graph application module enables graded detection of knowledge gaps and generation of personalized review paths. A teaching interaction module provides a multi-terminal adapted visual interactive interface, allowing students to view the knowledge graph and learning records, thereby improving teaching quality and learning efficiency.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an AI multi-scenario artificial intelligence education kit teaching system, which includes: a multimodal data acquisition module, a cross-modal semantic understanding module, a personal knowledge graph construction module, a knowledge graph application module, and a teaching interaction module; The multimodal data acquisition module connects to AI education kit hardware and various learning terminals to collect five types of multimodal raw data generated by students during the learning process: handwritten notes, voice explanations, code snippets, hardware operation videos, and experimental data. The cross-modal semantic understanding module receives multimodal raw data, extracts feature information from different modalities, and converts it into a unified semantic representation vector. The personal knowledge graph construction module receives a unified semantic representation vector, uses knowledge points as core nodes, learning resources, learning behaviors, and misconceptions as associated nodes, and mastery level, sequence, and causal relationship as edge attributes to construct and dynamically update the personal knowledge graph. The knowledge graph application module receives personal knowledge graph data and generates knowledge vulnerability detection results, personalized review paths, and visual data of learning outcomes. The teaching interaction module receives various types of data output by the knowledge graph application module, presents teaching content and visualized data, and receives and responds to students' interactive operations.

[0007] Furthermore, the multimodal data acquisition module includes a handwriting acquisition unit, a voice acquisition unit, a code acquisition unit, a video acquisition unit, and an experimental data acquisition unit: the handwriting acquisition unit connects to the electromagnetic induction handwriting tablet to acquire the student's handwriting trajectory coordinates, pressure values, and timestamp data; the voice acquisition unit connects to the microphone to acquire audio data; the code acquisition unit acquires character input, deletion, modification records, and code execution logs during the code editing process in real time through the integrated development environment; the video acquisition unit connects to the camera included with the educational kit to acquire hardware operation video data; the experimental data acquisition unit connects to various sensors in the educational kit to acquire raw physical quantity data during the experiment; all acquired data is accompanied by an acquisition timestamp and a unique data identifier and stored in a local cache.

[0008] Furthermore, the cross-modal semantic understanding module adopts a multimodal feature extraction architecture to extract feature information from different modalities of data: after handwritten data is converted into text content by OCR recognition, a text feature vector is generated through a word embedding layer, and handwriting semantic analysis is used to extract features such as writing speed, stroke order, and emphasis marking; after speech data is converted into text content by speech recognition, a text feature vector is generated, and speech rate, pause duration, and energy distribution features are extracted; after code data is analyzed by abstract syntax tree to extract code structure, function calls, and variable definition features; after video data is analyzed by action recognition and object detection to extract operation steps, hardware connection methods, and experimental phenomena features; after experimental data is analyzed by time series to extract data change trends, extreme points, and fluctuation features; all features extracted from all branches are converted into 768-dimensional vector representations.

[0009] Furthermore, the cross-modal semantic understanding module transforms the feature vectors of different modal data into a unified semantic representation vector, calculated using the following formula: ,in, The output is a unified semantic representation vector with a dimension of 768; It is the ReLU activation function; For handwritten data feature vectors; For speech data feature vectors; For code data feature vectors; For video data feature vectors; The feature vector of the experimental data; , , , , The fusion weighting coefficient is adjustable. This is the bias vector, with a dimension of 768. It is initially set to a zero vector and is adaptively adjusted through backpropagation as the system runs. , , , , The initial values ​​are 0.25, 0.2, 0.3, 0.15, and 0.1, respectively. During system operation, each weight coefficient will be adaptively adjusted according to the student's learning performance, with an adjustment step size of 0.01.

[0010] Furthermore, in the personal knowledge graph construction module: the core knowledge point nodes include five basic attributes: node ID, node name, node type, creation time, and last access time; the learning resource association nodes include four attributes: resource ID, resource type, resource path, and resource size; the learning behavior association nodes include four attributes: behavior ID, behavior type, behavior time, and behavior duration; the error cognition association nodes include four attributes: error ID, error type, error description, and number of occurrences; the mastery level edge attribute has a value range of 0 to 1; the sequence edge attribute represents the learning sequence relationship between two knowledge points; and the causal relationship edge attribute represents the logical causal relationship between two knowledge points.

[0011] Furthermore, the personal knowledge graph construction module determines the mastery level attribute value for each knowledge point node, calculated using the following formula: ,in, The attribute value represents the level of mastery of knowledge points, with a range of 0 to 1. The correct answer rate for the corresponding knowledge point ranges from 0 to 1. A = number of correct answers / total number of answers. The code execution success rate for the corresponding knowledge point ranges from 0 to 1, where B = number of successful code executions / total number of executions. The value ranges from 0 to 1, representing the quality of the experiment completion for the corresponding knowledge point. C = 1 - Normalized Absolute Deviation, which is calculated by the mean square error between the experimental data and the standard data. The number of days since the current time; 30 is a natural constant; 30 is a time constant, in days.

[0012] Furthermore, the personal knowledge graph construction module adopts an incremental update method, automatically triggering updates after each student completes a learning task, with the time interval between two consecutive updates not exceeding 10 minutes. The update process first obtains the newly generated unified semantic representation vector, then matches this vector with all existing nodes in the knowledge graph, and performs node association or creation operations based on the matching results; next, it updates the basic attributes of relevant nodes and the associated node information; finally, it adjusts the edge attribute values ​​between nodes, which are adjusted according to the number of times two nodes are accessed simultaneously and the amount of associated data, with higher edge attribute values ​​indicating a stronger association between the two nodes; after the update is completed, the latest knowledge graph is stored in the local database and synchronized to the cloud server.

[0013] Furthermore, the knowledge graph application module includes a knowledge vulnerability detection unit. This unit traverses all knowledge point nodes in the personal knowledge graph, extracting the mastery level attribute and erroneous cognitive association node information for each node. When the mastery level attribute value of a node is less than 0.6, the corresponding node is marked as a Level 1 knowledge vulnerability. When the mastery level attribute value of a node is greater than or equal to 0.6 and less than 0.7, and there is at least one erroneous cognitive association node, the corresponding node is marked as a Level 2 knowledge vulnerability. When the mastery level attribute value of a node is greater than or equal to 0.7 and less than 0.8, and there are at least two erroneous cognitive association nodes, the corresponding node is marked as a Level 3 knowledge vulnerability. In other cases, no knowledge vulnerability is marked. After detection, a knowledge vulnerability list is generated, sorted from high to low vulnerability level, and the error type and related learning records corresponding to each vulnerability are labeled.

[0014] Furthermore, the knowledge graph application module includes a review path generation unit. This unit generates personalized review paths based on the knowledge vulnerability list and the topology of the personal knowledge graph. Starting from the highest-level knowledge vulnerability node, it traverses all the prerequisite knowledge point nodes. If the mastery level attribute value of a prerequisite knowledge point node is lower than 0.7, the corresponding prerequisite node is added to the review path. Then, the review order is arranged from front to back according to the dependency relationship of the knowledge points. Each knowledge point node corresponds to a review task package, which includes micro-course videos, practice questions, and experimental projects. The difficulty of the review tasks is adjusted according to the mastery level attribute value of the knowledge point node. The lower the mastery level attribute value, the lower the task difficulty. After the review path is generated, it is displayed to students through the teaching interaction module.

[0015] Furthermore, the teaching interaction module provides a multi-terminal adapted human-computer interaction interface, supporting operation on tablets, computers, and mobile phones. The interface adopts a responsive design, which can automatically adjust the layout and element size according to the device screen size. The interaction interface includes a knowledge graph visualization area, a learning record display area, and a teaching content playback area. The knowledge graph visualization area displays the individual knowledge graph in the form of a force-directed graph, using different colors to represent the mastery level attribute value of nodes and using different thicknesses of edges to represent the edge attribute values ​​between nodes. Students can click on any node to view all learning records associated with the corresponding node, including handwritten notes, voice explanations, code snippets, hardware operation videos, and experimental data. At the same time, the layout of the knowledge graph can be adjusted by dragging and dropping.

[0016] Compared with existing technologies, this AI multi-scenario artificial intelligence education kit teaching system has the following beneficial effects: I. This invention integrates multimodal data acquisition and cross-modal semantic understanding technologies to achieve comprehensive collection and deep fusion of various types of data during students' learning process. The system interfaces with educational hardware and various learning terminals, covering multiple learning data types such as handwriting, voice, code, operation videos, and experimental data, fully restoring the entire learning process of students. By extracting feature information from different modal data and converting it into a unified semantic representation, it breaks down information barriers between different modal data and solves the problems of data isolation and one-sided analysis in existing technologies. It constructs a personal knowledge graph with knowledge points as the core, containing multiple related nodes and edge attributes, and uses an incremental update method to adjust the graph content in real time, which can dynamically and accurately depict students' knowledge mastery status and cognitive development trajectory.

[0017] Second, this invention provides students with precise and personalized teaching services through intelligent applications based on personal knowledge graphs and multi-terminal interactive design. The system can automatically traverse the knowledge point nodes in the knowledge graph, combine the level of mastery and cognitive errors to perform graded detection of knowledge gaps, and clearly locate students' weak knowledge areas. Based on knowledge gaps and the graph topology, it generates personalized review paths that conform to the dependencies between knowledge points and matches review tasks of corresponding difficulty to achieve individualized instruction. At the same time, it provides an interactive interface adapted to multiple terminals, displaying the personal knowledge graph in a visual form, allowing students to easily view various related learning records, significantly improving the intuitiveness and effectiveness of teaching interaction, and enhancing the teaching quality of the AI ​​education kit and students' learning efficiency.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 A block diagram showing the modular components of an AI multi-scenario artificial intelligence education kit teaching system; Figure 2 Flowchart for the construction and dynamic updating of personal knowledge graphs in an AI multi-scenario artificial intelligence education kit teaching system; Figure 3 A flowchart for knowledge gap detection and personalized review path generation in an AI multi-scenario artificial intelligence education kit teaching system. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] Example 1 The multimodal data acquisition module, serving as the system's data entry point, directly connects to the AI ​​education kit hardware and tablet learning terminal. It comprehensively collects five types of raw multimodal data generated by students during their AI basic programming experiments: handwritten notes, audio explanations, code snippets, hardware operation videos, and experimental data. This provides a complete and authentic data source for subsequent semantic analysis and knowledge modeling, preventing biases in learning assessments due to missing data. The handwriting acquisition unit specifically connects to the electromagnetic induction handwriting tablet, accurately capturing students' handwriting trajectory coordinates, pressure values, and timestamp data. This fully recreates the entire process of students writing programming notes, organizing algorithm logic, and marking key points and difficulties, clearly recording the students' thinking rhythm and knowledge organization logic reflected in their written work, transforming implicit learning thinking into quantifiable data. The audio acquisition unit connects to a professional microphone, comprehensively capturing audio data when students explain programming ideas, illustrate code functions, and summarize experimental problems. This preserves students' verbal expression and thought process, intuitively reflecting their oral understanding and logical organization abilities regarding programming knowledge points. The code acquisition unit uses an integrated development environment (IDE) to collect data in real time, recording the entire process of student code editing, including character input, deletion, modification, and code execution logs. It continuously tracks the student's programming trial-and-error, debugging, and optimization processes, providing a complete picture of the student's operation trajectory from code conception to functional implementation. The video acquisition unit connects to the high-definition camera included with the educational kit, continuously capturing videos of students' hardware setup, module connection, and functional debugging. It fully records the standardization of hardware operations and the integrity of the experimental process, capturing every operational action and change in experimental phenomena. The experimental data acquisition unit connects to various sensors on the educational kit, accurately collecting raw physical quantity data during the programming experiments. This accurately reflects the operating status and execution effect of the experimental device, providing a reliable basis for the objective evaluation of experimental results. All collected data is automatically accompanied by a timestamp and a unique data identifier, and stored uniformly in a local cache. This ensures the accurate temporal correspondence of different data types and guarantees secure data storage even offline, preventing loss or corruption due to network fluctuations. This ensures that every piece of learning data is traceable and unique.

[0023] The cross-modal semantic understanding module is responsible for receiving all the raw data transmitted by the multimodal data acquisition module. It employs a multimodal feature extraction architecture to process different modalities of data specifically, extracting the core feature information of each data type and uniformly converting it into a 768-dimensional vector representation. The formula is as follows: ,in, The output is a unified semantic representation vector with a dimension of 768; It is the ReLU activation function; For handwritten data feature vectors; For speech data feature vectors; For code data feature vectors; For video data feature vectors; The feature vector of the experimental data; , , , , The fusion weighting coefficient is adjustable. This is the bias vector, with a dimension of 768. It is initially set to a zero vector and is adaptively adjusted through backpropagation as the system runs. , , , , The initial values ​​are 0.25, 0.2, 0.3, 0.15, and 0.1, respectively. During system operation, each weight coefficient will be adaptively adjusted according to the student's learning performance, with an adjustment step size of 0.01. This eliminates format barriers between different modalities of data, laying a standardized foundation for subsequent information fusion. For handwritten data, OCR recognition is first used to convert handwritten content into standard text, and then a text feature vector is generated through a word embedding layer. At the same time, handwriting semantic analysis is combined to extract writing speed, stroke order, and key annotation features, which not only accurately restores the semantic information of the handwritten content but also reflects the student's focus and knowledge emphasis during writing. For speech data, speech recognition is first used to convert speech into text content and generate a text feature vector. At the same time, speech rate, pause duration, and energy distribution features are extracted to fully restore the core meaning of the speech expression, while reflecting the student's fluency of thought and confidence in knowledge mastery during explanation. For code data, abstract syntax tree analysis is used to extract code structure, function calls, and variable definition features, deeply analyzing the logical architecture and standardization of the code, identifying problems and highlights in the code, and intuitively presenting the student's programming ability level. For video data, action recognition and target detection are used to extract operation steps, hardware connection methods, and experimental phenomenon characteristics, automatically judging the correctness of the operation process and the validity of the experimental results, replacing manual observation to achieve accurate operation evaluation. For experimental data, time series analysis is used to extract data change trends, extreme points, and fluctuation characteristics, clearly presenting the changing patterns of experimental data and quickly identifying abnormal data and key nodes. After all feature extraction is completed, the module transforms the feature vectors of five types of modal data into a unified semantic representation vector through a cross-modal semantic fusion formula. Weights are adaptively allocated according to the information contribution of different modal data in programming learning, achieving the organic fusion of multi-source information and generating unified semantic data that comprehensively reflects students' programming learning status and knowledge understanding, transforming fragmented multimodal data into standardized information with holistic semantics.

[0024] The personal knowledge graph construction module receives a unified semantic representation vector output by the cross-modal semantic understanding module. Using AI programming-related knowledge points as core nodes, learning resources, learning behaviors, and misconceptions as associated nodes, and mastery level, sequence, and causal relationships as edge attributes, it builds a personalized programming knowledge graph for each student and dynamically updates it, clearly presenting the logical connections between knowledge points and the student's knowledge construction status. The core knowledge point nodes are configured with five basic attributes: node ID, node name, node type, creation time, and last access time, uniquely identifying each programming knowledge point, recording the learning time and review frequency, and comprehensively presenting the student's programming learning progress. The learning resource associated nodes include four attributes: resource ID, resource type, resource path, and resource size, enabling precise binding of learning resources to knowledge points for easy and quick access and matching of learning content. The learning behavior associated nodes include four attributes: behavior ID, behavior type, behavior time, and behavior duration, comprehensively recording each student's learning operation, reconstructing the learning behavior trajectory and time allocation. The error recognition association node includes four attributes: error ID, error type, error description, and frequency of occurrence. This accurately identifies students' knowledge gaps, operational errors, and logical flaws in programming learning, statistically analyzes error frequency, and pinpoints weaknesses. The module calculates the mastery level attribute value for each knowledge point node using a formula: ,in, The attribute value represents the level of mastery of knowledge points, with a range of 0 to 1. The correct answer rate for the corresponding knowledge point ranges from 0 to 1. A = number of correct answers / total number of answers. The code execution success rate for the corresponding knowledge point ranges from 0 to 1, where B = number of successful code executions / total number of executions. The value ranges from 0 to 1, representing the quality of the experiment completion for the corresponding knowledge point. C = 1 - Normalized Absolute Deviation, which is calculated by the mean square error between the experimental data and the standard data. The number of days since the current time; 30 is a natural constant; 30 is a time constant in days. The system comprehensively considers dimensions such as answer accuracy, code execution success rate, and experiment completion quality, combined with time decay factors, to objectively reflect students' real-time mastery of knowledge points, avoiding the one-sidedness of single-dimensional evaluation. The module adopts an incremental update method, automatically triggering an update after each student completes a programming learning task, with an interval of no more than 10 minutes between adjacent updates, ensuring the knowledge graph is always synchronized with the student's latest learning status. During the update, a new unified semantic representation vector is first obtained, matched with existing knowledge graph nodes to complete association or create a new one, then the node's basic attributes and association information are updated, and finally, edge attribute values ​​are adjusted based on the number of simultaneous visits to the node and the amount of associated data to strengthen the correlation between knowledge points. After the update is completed, the latest knowledge graph is synchronously stored in the local database and cloud server, ensuring offline use and multi-terminal data synchronization, while also achieving data security backup to prevent information loss.

[0025] The knowledge graph application module receives knowledge graph data transmitted from the personal knowledge graph construction module. Based on the structured knowledge data, it generates knowledge vulnerability detection results, personalized review paths, and visualized learning outcomes, transforming abstract knowledge structures into actionable learning guidance plans to provide precise direction for students' programming learning. The module's built-in knowledge vulnerability detection unit traverses all programming knowledge point nodes, extracts mastery level attributes and information related to misconceptions, and marks vulnerabilities according to a unified standard: nodes with a mastery level below 0.6 are marked as Level 1 knowledge vulnerabilities, indicating a serious lack of mastery and requiring priority reinforcement; nodes with a mastery level between 0.6 and 0.7 and at least one misconception node are marked as Level 2 knowledge vulnerabilities, indicating a clear cognitive misconception requiring targeted correction; nodes with a mastery level between 0.7 and 0.8 and at least two misconception nodes are marked as Level 3 knowledge vulnerabilities, indicating minor cognitive biases requiring remediation. After detection, a knowledge vulnerability list is generated, sorted from highest to lowest level, with each vulnerability's error type and associated learning record labeled, allowing students to clearly understand the distribution of weaknesses and learning priorities. The review path generation unit generates personalized review paths based on a vulnerability list and knowledge graph topology. Starting from the highest-level vulnerability node, it traverses the prerequisite knowledge points, including those with a mastery level below 0.7. The review path follows a hierarchical order of knowledge dependencies, progressing from simple to complex. Each knowledge point is matched with a review task package containing micro-course videos, practice questions, and experimental projects. The difficulty of the tasks is adaptively adjusted based on the student's mastery level; the lower the mastery level, the more suitable the task difficulty, helping students consolidate knowledge gradually and avoiding learning frustration due to inappropriate difficulty. The generated review path is directly transmitted to the teaching interaction module, providing students with a clear and actionable review plan.

[0026] The teaching interaction module receives test results, review paths, and visualization data from the knowledge graph application module. It is responsible for presenting teaching content, displaying data information, and responding to student interactions, building an efficient interactive bridge between students and the system, and providing a convenient and smooth learning experience. The module offers a multi-terminal compatible human-computer interface, supporting tablets, computers, and mobile phones. It adopts a responsive design, automatically adjusting the layout and element size to different screen sizes, ensuring a consistent user experience across all scenarios. The interface is divided into three main functional areas: a knowledge graph visualization area, a learning record display area, and a teaching content playback area, with clear division of labor and convenient operation. The knowledge graph visualization area displays a personal programming knowledge graph in the form of a force-directed graph, using different colors to distinguish the level of mastery of knowledge points and different thicknesses of edges to represent the strength of knowledge connections. Students can click on nodes to view related handwritten notes, audio explanations, code snippets, operation videos, and experimental data. They can also drag and drop to adjust the graph layout to meet personalized viewing needs. The learning record display area presents all learning activities in chronological order, clearly showing the learning timeline and progress, facilitating student summarization and reflection. The teaching content playback area supports micro-course video playback, online practice and answering, and experimental guidance demonstrations. It receives student actions in real time and provides feedback, creating an immersive programming learning environment. The entire interactive process is responsive and easy to operate, allowing students to intuitively grasp their learning status, efficiently conduct self-study and targeted review, and comprehensively improve the learning effect of basic artificial intelligence programming experiments. Figure 1 As shown.

[0027] Example 2 The multimodal data acquisition module, as the core of the system's data acquisition, connects to the AI ​​education kit hardware and computer learning terminal. It comprehensively collects five types of raw data generated by students during their hardware circuit design and debugging learning: handwritten circuit schematics and debugging notes, voice debugging explanations and problem analyses, code snippets, hardware soldering and connection operation videos, and circuit experiment data. This provides a comprehensive, realistic, and detailed data source for subsequent semantic analysis and knowledge modeling, fully covering information from all stages of circuit learning. The handwriting acquisition unit connects to an electromagnetic induction handwriting tablet, accurately collecting the handwriting trajectory coordinates, pressure values, and timestamp data of students drawing circuit schematics, labeling component parameters, and writing debugging notes. This fully reconstructs the conceptualization process of circuit design and the process of organizing debugging ideas, clearly recording students' thought processes regarding circuit structure, principles, and faults. The voice acquisition unit connects to a professional microphone, fully collecting audio data of students explaining circuit principles, analyzing fault causes, and summarizing debugging experience. This retains the students' verbal expressions and logical analysis processes, intuitively reflecting the depth of their understanding of circuit knowledge and their problem-solving abilities. The code acquisition unit uses the integrated development environment to collect real-time character input, deletion, and modification records and code execution logs of the microcontroller control program, tracking the entire process of student program writing, debugging, and burning, and fully presenting the programming operation trajectory of the embedded program and hardware circuit. The video acquisition unit connects to the educational kit's built-in camera, continuously capturing videos of students' electronic component soldering, circuit board assembly, circuit connection, and power-on testing operations. It fully records the standardization of hardware operations, soldering quality, and experimental procedures, capturing every operational detail and change in circuit phenomena. The experimental data acquisition unit connects to the educational kit's voltage, current, and resistance sensors, accurately collecting raw physical quantity data during circuit experiments. This accurately reflects the circuit's operating status, parameter changes, and debugging effects, providing precise data support for the objective evaluation of circuit experiments. All acquired data is automatically bound to a timestamp and a unique data identifier and stored uniformly in a local cache, ensuring accurate time-series correspondence and clear correlation of various data types. Offline storage is also supported to avoid data loss due to network interruptions, ensuring complete traceability of every operation and record in circuit learning.

[0028] The cross-modal semantic understanding module receives all the raw circuit learning data transmitted from the multimodal data acquisition module. It employs a multimodal feature extraction architecture to professionally process the five types of data, extracting core features and uniformly converting them into 768-dimensional vector representations. This overcomes format barriers between different modalities, providing a standardized data foundation for multi-source information fusion. For handwritten circuit schematics and debugging notes, OCR recognition is used to convert them into text content. A word embedding layer generates text feature vectors, and handwriting semantic analysis extracts features such as writing speed, stroke order, and key annotations. This accurately restores the textual semantics of circuit design and debugging while reflecting the student's focus and knowledge emphasis during drawing and writing. For audio explanations and fault analysis data, speech recognition is used to convert them into text and generate text feature vectors. Speech rate, pause duration, and energy distribution features are extracted to fully restore the core content of circuit principle explanations and fault analysis, reflecting the student's fluency of thought and confidence in their knowledge. For microcontroller code data, abstract syntax tree analysis is used to extract code structure, function calls, and variable definition features, deeply analyzing the logical architecture of program and circuit cooperation, identifying code vulnerabilities and standardization issues, and intuitively presenting students' embedded programming and circuit control capabilities. For hardware soldering and connection videos, action recognition and target detection are used to extract operation steps, connection methods, and experimental phenomena, automatically judging soldering quality, connection sequence, and circuit on / off status, accurately assessing the correctness of hardware operations. For circuit experiment data, time series analysis is used to extract data change trends, extreme points, and fluctuation characteristics, clearly presenting the changing patterns of parameters such as voltage and current, and quickly identifying circuit anomalies and critical nodes. After all feature extraction is completed, the module transforms five types of feature vectors into a unified semantic representation vector through a cross-modal semantic fusion formula. The fusion weights are adaptively allocated according to the information contribution of different modal data in circuit learning, achieving organic integration of handwritten, voice, code, video, and experimental data, generating unified semantic data that comprehensively reflects students' circuit design, soldering, debugging abilities, and knowledge mastery, transforming fragmented circuit learning data into standardized, integrated semantic information.

[0029] The personal knowledge graph construction module receives a unified semantic representation vector output by the cross-modal semantic understanding module. It uses circuit-related knowledge points such as circuit principles, component characteristics, soldering processes, program burning, and troubleshooting as core nodes, learning resources, learning behaviors, and misconceptions as associated nodes, and mastery level, sequence of learning, and causal relationships as edge attributes to construct a student-specific circuit knowledge graph that is dynamically updated in real time. This clearly presents the internal logic of circuit knowledge and the student's personal knowledge system construction. Core knowledge point nodes include five basic attributes: node ID, node name, node type, creation time, and last access time, uniquely identifying each circuit knowledge point and recording the learning time and review frequency, comprehensively presenting the student's circuit learning progress and frequency. Learning resource associated nodes are configured with four attributes: resource ID, resource type, resource path, and resource size, enabling precise binding of resources such as circuit teaching videos, schematic libraries, and debugging guides to knowledge points for convenient and quick access and matching. The learning behavior association nodes include four attributes: behavior ID, behavior type, behavior time, and behavior duration. They comprehensively record all student learning behaviors, including drawing diagrams, soldering operations, coding, and experimental debugging, reconstructing the learning trajectory and time allocation. Error cognition association nodes include four attributes: error ID, error type, error description, and frequency of occurrence. They accurately pinpoint students' knowledge gaps, operational errors, and logical flaws in circuit design, soldering, debugging, and programming, statistically analyzing error frequency and identifying weaknesses in circuit learning. The module calculates the mastery level attribute value for each circuit knowledge point using a formula that integrates three dimensions: correct answer rate, code execution success rate, and experimental completion quality, combined with time decay factors, to objectively reflect students' real-time mastery level of circuit knowledge points, avoiding the limitations of a single evaluation dimension. The module employs an incremental update method, automatically triggering updates after each student completes a circuit learning task, with adjacent updates occurring no more than 10 minutes apart, ensuring the knowledge graph is completely synchronized with the student's latest learning status. During the update process, a new unified semantic vector is first obtained, matched with existing nodes to complete associations or create new ones, then node attributes and association information are updated. Finally, edge attribute values ​​are adjusted based on the number of simultaneous visits to the node and the amount of associated data to enhance the realistic representation of the strength of knowledge associations. After the update is completed, the knowledge graph is synchronously stored in a local database and a cloud server to ensure normal offline use and multi-terminal data synchronization, while also achieving secure data backup to prevent the loss of circuit learning information, such as... Figure 2 As shown.

[0030] The knowledge graph application module receives circuit knowledge graph data from the personal knowledge graph construction module. Based on structured knowledge, it generates knowledge gap detection results, personalized review paths, and visualized learning outcomes, transforming the abstract circuit knowledge structure into a practical learning guidance plan, providing precise guidance for students' circuit learning. The module's knowledge gap detection unit traverses all circuit knowledge point nodes, extracts mastery levels and misconception information, and completes a level-based classification of gaps according to standards: nodes with a mastery level below 0.6 are marked as Level 1 knowledge gaps, indicating a severe lack of mastery of the circuit knowledge point, unable to support subsequent learning, requiring the highest priority for reinforcement; nodes with a mastery level between 0.6 and 0.7 and containing at least one misconception are marked as Level 2 knowledge gaps, indicating a clear cognitive misconception, requiring targeted correction; nodes with a mastery level between 0.7 and 0.8 and containing at least two misconception nodes are marked as Level 3 knowledge gaps, indicating multiple minor deviations, requiring further investigation and filling. After detection, a gap list is generated in descending order of level, annotating the error type and associated learning records, allowing students to clearly understand their weak points in circuit knowledge and their learning priorities. The review path generation unit generates personalized review paths based on a vulnerability list and knowledge graph topology. Starting from the highest-level vulnerability node, it traverses all prerequisite knowledge points, including those with a mastery level below 0.7. The review path strictly follows the dependency relationship of circuit knowledge, arranging the review in a progressively more complex order. Each knowledge point is matched with a review task package containing micro-course videos, practice questions, and hands-on debugging experiments. The difficulty of the tasks is adaptively adjusted according to the mastery level of the knowledge point; the lower the mastery level, the more suitable the difficulty, helping students gradually consolidate their circuit fundamentals and avoiding the negative impact of unsuitable difficulty on learning outcomes. The generated review path is transmitted to the teaching interaction module in real time, providing students with a clear and executable circuit review plan, such as... Figure 3 As shown.

[0031] The teaching interaction module receives vulnerability results, review paths, and visualization data from the knowledge graph application module. It is responsible for presenting circuit teaching content, displaying knowledge graph data, and responding to student interactions, building a convenient and efficient interactive channel for circuit learning and providing students with a smooth practical training experience. The module offers a multi-terminal compatible interface, supporting tablets, computers, and mobile phones. It adopts a responsive design, automatically adjusting the layout to different screen sizes to ensure a consistent user experience across all scenarios. The interactive interface is divided into three main sections: a knowledge graph visualization area, a learning record display area, and a teaching content playback area. The functions are clear and the operation is simple. The knowledge graph visualization area displays a personal circuit knowledge graph using a force-directed graph. Different colors distinguish the level of mastery of knowledge points, and different edge thicknesses indicate the strength of knowledge connections. Students can click on any node to view related circuit schematics, audio explanations, code snippets, soldering videos, and experimental data. They can also drag and drop to adjust the graph layout to meet personalized viewing needs. The learning record display area presents all learning activities, including circuit design, soldering, debugging, and programming, in chronological order, clearly displaying the learning timeline and progress, facilitating student summarization and reflection. The teaching content playback area supports smooth playback of circuit principle videos, soldering operation demonstrations, and debugging guidance tutorials. It supports online practice and answering questions, and viewing experimental steps. It receives student operations in real time and provides rapid feedback, creating an immersive hardware circuit learning environment. The entire interactive process is responsive and easy to operate, allowing students to intuitively grasp their own circuit knowledge mastery status, efficiently conduct independent training and targeted review, and comprehensively improve their learning efficiency and practical skills in hardware circuit design and debugging.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI multi-scenario artificial intelligence education kit teaching system, characterized in that: The system includes: a multimodal data acquisition module, a cross-modal semantic understanding module, a personal knowledge graph construction module, a knowledge graph application module, and a teaching interaction module; The multimodal data acquisition module connects to AI education kit hardware and various learning terminals to collect five types of multimodal raw data generated by students during the learning process: handwritten notes, voice explanations, code snippets, hardware operation videos, and experimental data. The cross-modal semantic understanding module receives multimodal raw data, extracts feature information from different modalities, and converts it into a unified semantic representation vector. The personal knowledge graph construction module receives a unified semantic representation vector, uses knowledge points as core nodes, learning resources, learning behaviors, and misconceptions as associated nodes, and mastery level, sequence, and causal relationship as edge attributes to construct and dynamically update the personal knowledge graph. The knowledge graph application module receives personal knowledge graph data and generates knowledge vulnerability detection results, personalized review paths, and visual data of learning outcomes. The teaching interaction module receives various types of data output by the knowledge graph application module, presents teaching content and visualized data, and receives and responds to students' interactive operations.

2. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The multimodal data acquisition module includes a handwriting acquisition unit, a voice acquisition unit, a code acquisition unit, a video acquisition unit, and an experimental data acquisition unit: the handwriting acquisition unit connects to the electromagnetic induction handwriting tablet to acquire the student's handwriting trajectory coordinates, pressure values, and timestamp data; the voice acquisition unit connects to the microphone to acquire audio data; the code acquisition unit acquires character input, deletion, modification records, and code execution logs in real time through the integrated development environment; the video acquisition unit connects to the camera included with the educational kit to acquire hardware operation video data; the experimental data acquisition unit connects to various sensors in the educational kit to acquire raw physical quantity data during the experiment; all acquired data is accompanied by an acquisition timestamp and a unique data identifier and stored in a local cache.

3. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The cross-modal semantic understanding module adopts a multimodal feature extraction architecture to extract feature information from different modal data: after handwritten data is converted into text content by OCR recognition, text feature vectors are generated through word embedding layer, and writing speed, stroke order and key annotation features are extracted through handwriting semantic analysis. After speech data is converted into text content through speech recognition, a text feature vector is generated, and features such as speech rate, pause duration, and energy distribution are extracted. Code data is analyzed through abstract syntax tree to extract features such as code structure, function calls, and variable definitions. Video data is processed through action recognition and target detection to extract operational steps, hardware connection methods, and experimental phenomena characteristics; experimental data is processed through time series analysis to extract data change trends, extreme points, and fluctuation characteristics. The features extracted from all branches are converted into 768-dimensional vector representations.

4. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The cross-modal semantic understanding module transforms the feature vectors of different modal data into a unified semantic representation vector, calculated using the following formula: ,in, The output is a unified semantic representation vector with a dimension of 768; It is the ReLU activation function; For handwritten data feature vectors; For speech data feature vectors; For code data feature vectors; For video data feature vectors; The feature vector of the experimental data; , , , , The fusion weighting coefficient is adjustable. This is the bias vector, with a dimension of 768. It is initially set to a zero vector and is adaptively adjusted through backpropagation as the system runs. , , , , The initial values ​​are 0.25, 0.2, 0.3, 0.15, and 0.1, respectively. During system operation, each weight coefficient will be adaptively adjusted according to the student's learning performance, with an adjustment step size of 0.

01.

5. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, In the personal knowledge graph construction module: the core knowledge point node includes five basic attributes: node ID, node name, node type, creation time, and last access time; the learning resource association node includes four attributes: resource ID, resource type, resource path, and resource size; and the learning behavior association node includes four attributes: behavior ID, behavior type, behavior time, and behavior duration. The error cognition association node includes four attributes: error ID, error type, error description, and number of occurrences; the mastery level edge attribute has a value range of 0 to 1; the sequence edge attribute represents the learning sequence relationship between two knowledge points; and the causal relationship edge attribute represents the logical causal relationship between two knowledge points.

6. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The personal knowledge graph construction module determines the mastery level attribute value for each knowledge point node, and the calculation formula is as follows: ,in, The attribute value represents the level of mastery of knowledge points, with a range of 0 to 1. The correct answer rate for the corresponding knowledge point ranges from 0 to 1. A = number of correct answers / total number of answers. The code execution success rate for the corresponding knowledge point ranges from 0 to 1, where B = number of successful code executions / total number of executions. The value ranges from 0 to 1, representing the experimental completion quality for the corresponding knowledge point. C = 1 - Normalized Absolute Deviation, which is calculated using the mean square error between the experimental data and the standard data. The number of days since the current time; 30 is a natural constant; 30 is a time constant, in days.

7. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The personal knowledge graph construction module adopts an incremental update method, which automatically triggers an update after each student completes a learning task, and the time interval between two consecutive updates does not exceed 10 minutes. The update process first obtains the newly generated unified semantic representation vector, then matches the vector with all existing nodes in the knowledge graph, and performs node association or creation operations based on the matching results; then it updates the basic attributes of the relevant nodes and the associated node information; finally, it adjusts the edge attribute values ​​between nodes. The edge attribute values ​​are adjusted according to the number of times the two nodes are accessed simultaneously and the amount of associated data. The higher the edge attribute value, the stronger the association between the two nodes. After the update is complete, the latest knowledge graph will be stored in the local database and synchronized to the cloud server.

8. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The knowledge graph application module includes a knowledge vulnerability detection unit, which traverses all knowledge point nodes in the personal knowledge graph and extracts the mastery level attribute and erroneous cognition related node information for each node. When a node's mastery level attribute value is less than 0.6, the corresponding node is marked as a Level 1 knowledge vulnerability; when a node's mastery level attribute value is greater than or equal to 0.6 and less than 0.7, and there is at least one related node with incorrect cognition, the corresponding node is marked as a Level 2 knowledge vulnerability; when a node's mastery level attribute value is greater than or equal to 0.7 and less than 0.8, and there are at least two related nodes with incorrect cognition, the corresponding node is marked as a Level 3 knowledge vulnerability; otherwise, it is not marked as a knowledge vulnerability. After the detection is completed, a knowledge vulnerability list is generated, sorted from high to low vulnerability level, and the error type and related learning records for each vulnerability are labeled.

9. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The knowledge graph application module includes a review path generation unit. This unit generates personalized review paths based on the knowledge vulnerability list and the topology of the individual knowledge graph. Starting from the highest-level knowledge vulnerability node, it traverses all the prerequisite knowledge point nodes. If the mastery level attribute value of a prerequisite knowledge point node is lower than 0.7, the corresponding prerequisite node is added to the review path. Then, the review order is arranged from front to back according to the dependency relationship of the knowledge points. Each knowledge point node corresponds to a review task package, which includes micro-course videos, practice questions, and experimental projects. The difficulty of the review tasks is adjusted according to the mastery level attribute value of the knowledge point node. The lower the mastery level attribute value, the lower the task difficulty. After the review path is generated, it is displayed to students through the teaching interaction module.

10. The AI ​​multi-scenario artificial intelligence education kit teaching system according to claim 1, characterized in that, The teaching interaction module provides a multi-terminal adapted human-computer interaction interface, supporting operation on tablets, computers, and mobile phones. The interface adopts a responsive design, which can automatically adjust the layout and element size according to the device screen size. The interaction interface includes a knowledge graph visualization area, a learning record display area, and a teaching content playback area. The knowledge graph visualization area displays the individual knowledge graph in the form of a force-directed graph, using different colors to represent the mastery level attribute value of nodes and using different thicknesses of edges to represent the edge attribute values ​​between nodes. Students can click on any node to view all learning records associated with the corresponding node, including handwritten notes, voice explanations, code snippets, hardware operation videos, and experimental data. At the same time, the layout of the knowledge graph can be adjusted by dragging and dropping.

Images