Multi-modal artificial intelligence learning assistant based on embedded platform

Abstract

The invention belongs to the technical field of intelligent education, and discloses a multi-modal artificial intelligence learning assistant based on an embedded platform, and the method comprises the steps: a data collection and processing module receives learning problems and student learning data, and carries out the standardization processing, and obtains a problem standardized data set; the cognitive model construction module constructs a learner cognitive model based on the problem standardized data set and student learning data, and marks knowledge gap nodes; the teaching scheme generation module determines a self-adaptive teaching scheme including knowledge depth parameters, a learning material combination strategy and an explanation expression strategy according to the structural features of the cognitive model; the learning sequence construction module constructs a customized learning sequence according to the knowledge gap nodes and the knowledge depth parameters; and the learning content generation module generates personalized learning content according to the learning sequence and the teaching scheme. Through the self-adaptive processing flow driven by the cognitive model, personalized teaching for different students is realized, and the learning efficiency and the learning effect are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and more specifically, to a multimodal artificial intelligence learning assistant based on an embedded platform. Background Art

[0002] With the rapid development of artificial intelligence (AI), intelligent education systems have become a key research direction and application tool in the education field. Driven in particular by technologies such as natural language processing and knowledge graphs, AI-based intelligent teaching assistance systems can provide students with immediate learning support and knowledge answers, greatly expanding the boundaries and possibilities of traditional education. These systems, by understanding students' questions and generating relevant learning materials and answers, are playing an increasingly important role in various educational scenarios. However, existing intelligent teaching systems still have significant shortcomings in personalized learning support, limiting their teaching effectiveness and application value.

[0003] Most current AI-assisted teaching systems employ a simple "one question, one answer" model, whereby the system directly generates standardized responses or learning materials for specific student-generated questions. While technically simple, this approach ignores individual differences in the learning process and fails to adapt to the cognitive characteristics and learning needs of different students. Specifically, when different students input the same learning question, the system returns identical learning content, disregarding differences in knowledge base, learning styles, and cognitive abilities. This solution clearly contradicts the concept of personalized education and fails to meet the deeper needs of intelligent education. Summary of the Invention

[0004] In order to overcome the problem that existing technologies are difficult to meet the deep-seated needs of intelligent education, the present invention proposes a multimodal artificial intelligence learning assistant based on an embedded platform to solve the above problem.

[0005] The present invention provides the following technical solutions: A multimodal artificial intelligence learning assistant based on an embedded platform, comprising: The data collection and processing module is used to receive learning questions and student learning data, and standardize the learning questions to obtain a standardized data set of questions; Cognitive model building module, which is used to build a learner cognitive model based on the standardized problem dataset and student learning data, and mark the knowledge gap nodes in the learner cognitive model; A teaching plan generation module is used to determine an adaptive teaching plan based on the structural characteristics of the learner's cognitive model. The adaptive teaching plan includes a knowledge depth parameter, a learning material combination strategy, and an explanation and expression strategy. A learning sequence building module is used to build customized learning sequences based on knowledge gap nodes and knowledge depth parameters; The learning content generation module is used to generate learning content based on customized learning sequences, learning material combination strategies and explanation expression strategies.

[0006] Preferably, the step of standardizing the learning problem to obtain a standardized problem dataset includes: Convert the learning question into question text, clean the question text, remove extra spaces and special characters, and obtain the cleaned question text; Perform word segmentation on the cleaned question text to divide the question text into semantic units, and identify subject entities, skill keywords, and operation points from the semantic units; Classify subject entities, skill keywords, and operation points into professional fields, corresponding to specific training modules and knowledge points; Integrate the cleaned question text, semantic units, subject entities, skill keywords, operation points and professional field classification results to generate a standardized question dataset.

[0007] Preferably, the constructing of a learner cognitive model based on the standardized problem dataset and the student learning data includes: Extract student learning data, including relevant theoretical course grades, historical practical training scores, learning behavior records, and skill mastery levels; The subject entities, skill keywords and operation points identified in the standardized problem dataset are used as nodes to build an initial cognitive model; Based on the professional field classification results, the field association relationship between nodes is established and the field association relationship is mapped into the edge in the initial cognitive model; By analyzing students' theoretical course scores and skill mastery, we calculate students' mastery of each knowledge node and add mastery attributes to the nodes in the initial cognitive model. Based on students' historical learning behaviors and practical training scores, the students' familiarity with the knowledge association paths is calculated, and familiarity weight values ​​are assigned to the edges in the initial cognitive model to obtain the learner cognitive model.

[0008] Preferably, the calculation of the student's mastery of each knowledge node by analyzing the student's theoretical course performance and skill mastery level includes: Map each node in the initial cognitive model to the corresponding theoretical course knowledge points and practical training skill points; Extract students' historical performance data in corresponding theoretical courses and standardize the performance data to obtain theoretical mastery scores; Extract the scoring data of students' relevant skill points in history training, and standardize the scoring data to obtain the skill mastery score; The mastery degree is obtained by weighted summing the theoretical mastery score and the skill mastery score according to the preset weights.

[0009] Preferably, the calculating of the student's familiarity with the knowledge association path based on the student's historical learning behavior and practical training score includes: Analyze students' historical learning behavior data, including learning material access records, practical training operation sequences, and problem-solving paths; For each node pair connected by an edge in the initial cognitive model, extract the student's behavior sequence involving the node pair in historical learning; Calculate the average operation time and accuracy of students when solving problems related to this node; Familiarity was obtained based on average operation time and accuracy.

[0010] Preferably, marking the knowledge gap nodes in the learner's cognitive model includes: Identify all nodes in the learner's cognitive model whose mastery attribute value is below a preset threshold; Obtain the connection relationship of these low-mastery nodes in the cognitive model and construct a knowledge gap subgraph; Calculate the importance score of each node based on its degree; Sort the nodes in the knowledge gap subgraph according to their importance scores, and select the first preset number of nodes with the highest scores as target nodes; Analyze the pre-knowledge dependencies of the target node and identify the basic knowledge points that must be mastered first; Mark the target node and necessary prerequisite knowledge points as knowledge gap nodes.

[0011] Preferably, determining the adaptive teaching plan according to the structural characteristics of the learner's cognitive model includes: The knowledge coverage coefficient is calculated based on the ratio of the number of knowledge gap nodes to the total number of cognitive model nodes; the knowledge depth coefficient is calculated based on the average in-and-out degree and association strength of the knowledge gap nodes; The knowledge coverage breadth coefficient and the knowledge depth coefficient are combined to form the knowledge depth parameter; Calculate the average time students spend on each type of learning material and the completion rate, normalize the time and completion rate, and then take the weighted sum to obtain the preference score for each type of learning material; According to the preference scores of various learning materials, weight values ​​are allocated proportionally, and the sum of the weights is 1 to form a learning material combination strategy; The complexity of content expression is calculated based on the average score of students' history grades and the technical complexity of knowledge gap nodes. Based on the complexity of content expression, the maximum proportion of professional terms in the total vocabulary and the depth of the level of explanation that must be given to the concepts are determined to form an explanation expression strategy; A complete adaptive teaching plan is formed by integrating the three dimensions of knowledge depth parameters, learning material combination strategies and explanation and expression strategies.

[0012] Preferably, the step of constructing a customized learning sequence based on knowledge gap nodes and knowledge depth parameters includes: Based on the relationship between the previous learning requirements of the knowledge gap nodes, a learning dependency directed graph is constructed, where the direction of the edge indicates the direction from the previous knowledge to the subsequent knowledge; Apply the topological sorting algorithm to process the learning dependency directed graph and generate a knowledge point learning sequence that satisfies the principle of prioritizing learning of prior knowledge. Based on the knowledge depth parameter in the adaptive teaching plan, the number of surrounding knowledge points N that each knowledge gap node needs to include is calculated. The N surrounding knowledge points with the lowest familiarity with each gap node are selected from the cognitive model and added to the learning path as auxiliary learning content. Determine the number of content levels M for each knowledge point based on the knowledge depth coefficient, and plan M levels of content with increasing difficulty for each knowledge point, including basic concepts, principle analysis, application methods, problem solving, and advanced expansion, to form a vertical learning structure for knowledge points; The learning sequence generated by topological sorting is used as the backbone, combined with the N surrounding knowledge points of each node to form a horizontal expansion network, and the M content levels of each node to form a vertical depth structure, forming a complete customized learning sequence.

[0013] Preferably, the generation of learning content according to the customized learning sequence, learning material combination strategy and explanation expression strategy includes: Generate content for each level of each node in the customized learning sequence according to the content control strategy; The content control strategy includes: controlling the proportion of various types of learning materials in the generated content according to the learning material combination strategy, and controlling the number of professional terms and corresponding concept explanations in the generated content according to the explanation expression strategy.

[0014] The present invention provides a multimodal artificial intelligence learning assistant based on an embedded platform, which has the following beneficial effects: By standardizing learning problems through the Data Collection and Processing module and analyzing student learning data through the Cognitive Model Building module, the system can accurately construct personalized cognitive models that reflect students' knowledge status and accurately identify knowledge gaps. This dual modeling approach, based on deep semantic analysis and learning behavior data, enables the system to fully grasp students' cognitive characteristics and learning needs, laying a solid foundation for subsequent personalized instruction.

[0015] Based on the learner's cognitive model, the teaching plan generation module can develop adaptive teaching plans based on three dimensions: knowledge depth, learning material combination, and explanation and expression. This plan guides the learning sequence construction module to create learning paths that are both consistent with knowledge logic and adapted to individual characteristics, and guides the learning content generation module to provide personalized learning materials with diverse forms and appropriate expression. This integrated personalized approach, from teaching strategies to learning paths to specific content, ensures the consistency and targeted nature of the learning process. Through the adaptive learning system driven by cognitive models, each student can be provided with a tailored learning experience. This personalized approach significantly improves learning efficiency, reduces cognitive load, and enhances learning motivation, thereby comprehensively improving learning outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a module diagram of a multimodal artificial intelligence learning assistant based on an embedded platform of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example

[0018] See also Figure 1 In this embodiment, a multimodal artificial intelligence learning assistant based on an embedded platform includes: The data collection and processing module is used to receive learning questions and student learning data, and standardize the learning questions to obtain a standardized data set of questions; Standardized datasets obtained by standardizing learning problems include: Convert the learning question into question text, clean the question text, remove extra spaces and special characters, and obtain the cleaned question text; Perform word segmentation on the cleaned question text to divide the question text into semantic units, and identify subject entities, skill keywords, and operation points from the semantic units; Classify subject entities, skill keywords, and operation points into professional fields, corresponding to specific training modules and knowledge points; Integrate the cleaned question text, semantic units, subject entities, skill keywords, operation points and professional field classification results to generate a standardized question dataset.

[0019] In this embodiment, the data collection and processing module receives learning questions and student learning data and performs standardization processing.

[0020] First, learning questions and student history data can be received through a web interface or database connection. For the received learning questions, they are first converted into a unified text format and then cleaned up. For example, "How to implement the I2C communication protocol on an STM32 microcontroller?!" can be cleaned up into "How to implement the I2C communication protocol on an STM32 microcontroller?" The cleaned text enters the word segmentation stage, where a hybrid word segmentation strategy is used to segment the text into semantic units. From these units, the system identifies three types of key information: subject entities (such as "STM32" and "I2C"), skill keywords (such as "implementation"), and operational points (such as "how").

[0021] Next, the system classifies this information into professional fields, determines the subject area to which the problem belongs (such as "embedded system development"), and maps it to specific training modules (such as "embedded communication protocol implementation") and knowledge points (such as "STM32 hardware I2C configuration" and "I2C protocol timing").

[0022] Finally, we integrate all information to generate a standardized dataset for the problem, including cleaned text, word segmentation results, identified entities and keywords, and professional classification results. This standardized dataset provides foundational support for subsequent cognitive model construction, enabling intelligent understanding and processing of learning problems.

[0023] Cognitive model building module, which is used to build a learner cognitive model based on the standardized problem dataset and student learning data, and mark the knowledge gap nodes in the learner cognitive model; Building a learner cognitive model based on the standardized dataset of questions and student learning data includes: Extract student learning data, including relevant theoretical course grades, historical practical training scores, learning behavior records, and skill mastery levels; The subject entities, skill keywords and operation points identified in the standardized problem dataset are used as nodes to build an initial cognitive model; Based on the professional field classification results, the field association relationship between nodes is established and the field association relationship is mapped into the edge in the initial cognitive model; By analyzing students' theoretical course scores and skill mastery, we calculate students' mastery of each knowledge node and add mastery attributes to the nodes in the initial cognitive model. Based on students' historical learning behaviors and practical training scores, the students' familiarity with the knowledge association paths is calculated, and familiarity weight values ​​are assigned to the edges in the initial cognitive model to obtain the learner cognitive model.

[0024] By analyzing students' theoretical course scores and skill mastery, we calculate students' mastery of each knowledge node, including: Map each node in the initial cognitive model to the corresponding theoretical course knowledge points and practical training skill points; Extract students' historical performance data in corresponding theoretical courses and standardize the performance data to obtain theoretical mastery scores; Extract the scoring data of students' relevant skill points in history training, and standardize the scoring data to obtain the skill mastery score; The mastery degree is obtained by weighted summing the theoretical mastery score and the skill mastery score according to the preset weights.

[0025] Based on students' historical learning behaviors and practical training scores, the calculation of students' familiarity with knowledge association paths includes: Analyze students' historical learning behavior data, including learning material access records, practical training operation sequences, and problem-solving paths; For each node pair connected by an edge in the initial cognitive model, extract the student's behavior sequence involving the node pair in historical learning; Calculate the average operation time and accuracy of students when solving problems related to this node; Familiarity was obtained based on average operation time and accuracy.

[0026] The nodes that mark the knowledge gaps in the learner's cognitive model include: Identify all nodes in the learner's cognitive model whose mastery attribute value is below a preset threshold; Obtain the connection relationship of these low-mastery nodes in the cognitive model and construct a knowledge gap subgraph; Calculate the importance score of each node based on the node's in-degree; Sort the nodes in the knowledge gap subgraph according to their importance scores, and select the first preset number of nodes with the highest scores as target nodes; Analyze the pre-knowledge dependencies of the target node and identify the basic knowledge points that must be mastered first; Mark the target node and necessary prerequisite knowledge points as knowledge gap nodes.

[0027] In this embodiment, the cognitive model construction module constructs a learner cognitive model based on the standardized problem data set and the student learning data, and marks the knowledge gap nodes in the learner cognitive model.

[0028] First, student learning data can be extracted, including relevant theoretical course grades, historical practical training scores, learning behavior records and skill mastery levels.

[0029] Next, we can use the subject entities, skill keywords, and key operational points identified in the standardized problem dataset as nodes to construct an initial cognitive model. For example, for the aforementioned problem of "Implementing the I2C communication protocol on an STM32 microcontroller," we can use "STM32," "I2C," "communication protocol," and "implementation" as nodes in the cognitive model. These nodes represent the knowledge and skills that learners need to master and serve as the fundamental building blocks for constructing the cognitive model.

[0030] Based on the domain classification results, we can establish domain associations between nodes and map these associations as edges in the initial cognitive model. For example, in the embedded systems domain, there is a "hardware interface support" association between "STM32" and "I2C," and a "protocol type" association between "I2C" and "communication protocol." These associations reflect the logical connections between concepts in the knowledge system and help build a structured cognitive model.

[0031] By analyzing students' theoretical course grades and skill mastery, we can calculate their mastery of each knowledge node and add mastery attributes to the nodes in the initial cognitive model. Mastery is a quantitative indicator that measures students' understanding and application of specific knowledge points, directly influencing the development of subsequent personalized teaching plans. The calculation process involves mapping each node in the cognitive model to the corresponding theoretical course knowledge point and practical training skill point; extracting students' historical performance data for the corresponding theoretical course and standardizing the performance data to obtain a theoretical mastery score; extracting students' historical practical training scores for the relevant skill points and standardizing the scores to obtain a skill mastery score; and finally, taking the weighted sum of the theoretical and skill mastery scores according to preset weights to obtain the mastery score. For example, the "I2C Communication Protocol" node can be mapped to the "I2C Protocol" section in the "Embedded Communication Principles" course and the "I2C Communication Implementation" skill point in the "Embedded Communication Practice" course. If a student scores 85 points (normalized to 0.85) on the chapter test and 70 points (normalized to 0.7) on the practical skills assessment, with theory and practice weighted at 0.4 and 0.6, respectively, then the student's mastery of the "I2C Communication Protocol" node is 0.76. This mastery reflects the student's comprehensive understanding of the I2C protocol and provides a quantitative basis for personalized adjustments to subsequent teaching plans.

[0032] Based on students' historical learning behavior and practical training scores, we can calculate their familiarity with knowledge-related paths and assign familiarity weights to the edges in the initial cognitive model, resulting in a complete learner cognitive model. Familiarity reflects students' ability to apply knowledge points in a coordinated manner. It is an important auxiliary indicator for assessing students' mastery of knowledge points and reflects the possible reasons for their mastery. The calculation method involves analyzing students' historical learning behavior data; for each node pair connected by an edge in the model, extracting the student's behavioral sequences involving that node pair in their historical learning; calculating the student's average operation time and accuracy rate when solving related problems; and constructing a behavioral effectiveness vector based on this data to calculate familiarity. For example, for the edge connecting "STM32" and "I2C," we can analyze the student's operation records when implementing the STM32's I2C communication function. If a student takes an average of 15 minutes to complete the relevant configuration (the standard time is 20 minutes), the normalized time is 1, the normalized accuracy rate is 80% is 0.8, and the familiarity can be calculated as 1×0.8. The higher the familiarity value, the stronger the student's ability to apply the association between the two knowledge points, which helps the system judge the completeness of the student's knowledge structure.

[0033] After building a complete learner cognitive model, knowledge gap nodes can be marked. These nodes serve as the focus of subsequent personalized instruction. The marking process first identifies all nodes whose mastery level falls below a preset threshold (e.g., 0.6). The connections between these nodes are then constructed into a knowledge gap subgraph. The importance score of each node in the subgraph is calculated. Importance is calculated using the node's degree, with higher degrees indicating higher importance. The node with the highest importance is selected as the target node. The target node's dependencies on pre-requisite knowledge are analyzed. Finally, the target node and the necessary pre-requisite knowledge points are marked as knowledge gap nodes.

[0034] The cognitive model constructed in this way not only reflects students' mastery of individual knowledge points but also demonstrates their ability to apply them across them, providing a comprehensive assessment of their knowledge status. Furthermore, the cognitive model clearly identifies weak links in learning, providing precise targeting for subsequent instructional plan generation modules.

[0035] The teaching plan generation module is used to determine the adaptive teaching plan based on the structural characteristics of the learner's cognitive model. The adaptive teaching plan includes knowledge depth parameters, learning material combination strategies, and explanation and expression strategies; Determining adaptive teaching plans based on the structural characteristics of learners' cognitive models includes: The knowledge coverage coefficient is calculated based on the ratio of the number of knowledge gap nodes to the total number of cognitive model nodes; the knowledge depth coefficient is calculated based on the average in-and-out degree and association strength of the knowledge gap nodes; The knowledge coverage breadth coefficient and the knowledge depth coefficient are combined to form the knowledge depth parameter; Calculate the average time students spend on each type of learning material and the completion rate, normalize the time and completion rate, and then take the weighted sum to obtain the preference score for each type of learning material; According to the preference scores of various learning materials, weight values ​​are allocated proportionally, and the sum of the weights is 1 to form a learning material combination strategy; The complexity of content expression is calculated based on the average score of students' history grades and the technical complexity of knowledge gap nodes. Based on the complexity of content expression, the maximum proportion of professional terms in the total vocabulary and the depth of the level of explanation that must be given to the concepts are determined to form an explanation expression strategy; A complete adaptive teaching plan is formed by integrating the three dimensions of knowledge depth parameters, learning material combination strategies and explanation and expression strategies.

[0036] In this embodiment, the teaching plan generation module determines an adaptive teaching plan according to the structural characteristics of the learner's cognitive model, including a knowledge depth parameter, a learning material combination strategy, and an explanation and expression strategy.

[0037] First, the knowledge depth parameter can be calculated based on the characteristics of the knowledge gap nodes. The knowledge depth parameter is composed of the knowledge coverage breadth coefficient and the knowledge depth coefficient, reflecting the scope and depth of knowledge coverage that should be adopted in teaching. The knowledge coverage breadth coefficient can be calculated by the ratio of the number of knowledge gap nodes to the total number of nodes in the cognitive model. For example, if a student's cognitive model contains a total of 50 nodes, of which 15 are marked as knowledge gap nodes, the knowledge coverage breadth coefficient is 15 / 50 = 0.3. This coefficient reflects the extent of the gap in the student's knowledge system. The higher the coefficient, the wider the gap, and the more knowledge points the teaching plan needs to cover.

[0038] The knowledge depth coefficient can be calculated based on the average in-and-out degree and association strength of knowledge gap nodes. This coefficient reflects the complexity and importance of knowledge gap nodes in the knowledge network. A higher coefficient indicates that these nodes have more complex connections with other knowledge points and require more in-depth teaching.

[0039] The knowledge coverage breadth coefficient and the knowledge depth coefficient can be combined to form the knowledge depth parameter. The knowledge depth parameter directly affects the structure of the subsequent learning sequence and determines the balance between the breadth and depth of the learning content.

[0040] Next, we can calculate the average student dwell time and completion rate for each type of learning material to determine the learning material's preference score. For example, for four types of learning material: video tutorials, interactive experiments, text tutorials, and diagrams, we can collect historical student learning data. Assuming a student's average dwell time on a video tutorial is 25 minutes (normalized to 0.83) and their completion rate is 90% (normalized to 0.9), the video tutorial's preference score would be 0.83 × 0.5 + 0.9 × 0.5 = 0.865. Similarly, we can calculate the preference scores for other types of learning material: 0.92 for interactive experiments, 0.75 for text tutorials, and 0.68 for diagrams.

[0041] Based on the preference scores of various learning materials, weight values ​​can be assigned proportionally, and the sum of the weights can be made to 1, forming a learning material combination strategy. These weight values ​​constitute the learning material combination strategy, which guides the system to allocate the proportion of various materials when generating learning content.

[0042] The content complexity can be calculated based on the student's average history score and the technical complexity level of the knowledge gap nodes. For example, if a student's average history score is 82 (out of 100) and the average technical complexity of the knowledge gap nodes is 0.75 (out of 1), the content complexity can be calculated as (82 / 100) × 0.6 + (0.75) × 0.4 = 0.792. Based on this complexity, the maximum proportion of professional terms in the total vocabulary and the depth of concept explanation can be determined. For example, when the content complexity is 0.792, the maximum proportion of professional terms can be set to 30%, and the depth of concept explanation can be set to 2 (that is, professional concepts need to be explained using more basic concepts, but these basic concepts do not require further explanation).

[0043] Finally, a comprehensive adaptive teaching plan can be formed by integrating the three dimensions of knowledge depth parameters, learning material combination strategies, and explanation and presentation strategies. This teaching plan serves as a guiding framework for the subsequent construction of learning sequences and the generation of learning content, ensuring that the teaching content and format are aligned with students' cognitive states and learning preferences.

[0044] The adaptive teaching plan generated in this way not only takes into account the current status and gaps of students' knowledge mastery, but also makes full use of students' learning behavior data to achieve personalized customization of teaching content and form.

[0045] A learning sequence building module is used to build customized learning sequences based on knowledge gap nodes and knowledge depth parameters; Building customized learning sequences based on knowledge gap nodes and knowledge depth parameters includes: Based on the relationship between the previous learning requirements of the knowledge gap nodes, a learning dependency directed graph is constructed, where the direction of the edge indicates the direction from the previous knowledge to the subsequent knowledge; Apply the topological sorting algorithm to process the learning dependency directed graph and generate a knowledge point learning sequence that satisfies the principle of prioritizing learning of prior knowledge. Based on the knowledge depth parameter in the adaptive teaching plan, the number of surrounding knowledge points N that each knowledge gap node needs to include is calculated. The N surrounding knowledge points with the lowest familiarity with each gap node are selected from the cognitive model and added to the learning path as auxiliary learning content. Determine the number of content levels M for each knowledge point based on the knowledge depth coefficient, and plan M levels of content with increasing difficulty for each knowledge point, including basic concepts, principle analysis, application methods, problem solving, and advanced expansion, to form a vertical learning structure for knowledge points; The learning sequence generated by topological sorting is used as the backbone, combined with the N surrounding knowledge points of each node to form a horizontal expansion network, and the M content levels of each node to form a vertical depth structure, forming a complete customized learning sequence.

[0046] In this embodiment, the learning sequence construction module constructs a customized learning sequence based on the knowledge gap nodes and knowledge depth parameters to provide students with a personalized learning path.

[0047] First, a directed learning dependency graph can be constructed based on the relationships between knowledge gap nodes and the prerequisite learning requirements. In a directed graph, the direction of the edges indicates the direction from the prerequisite knowledge to the subsequent knowledge, reflecting the learning dependencies between knowledge points. For example, for the implementation of the I2C communication protocol in embedded systems learning, the following dependency relationship can be constructed: first learn "I2C basic concepts," then learn "I2C communication principles," then learn "STM32 hardware I2C interface," and finally learn "I2C communication program implementation." This dependency relationship ensures logical coherence in the learning process and prevents students from learning advanced content without the necessary prerequisite knowledge.

[0048] After constructing the directed learning dependency graph, a topological sorting algorithm can be applied to it to generate a learning sequence for each knowledge point that prioritizes prerequisite knowledge. Topological sorting ensures that all prerequisite knowledge for any knowledge point is prioritized within the learning sequence. For example, topologically sorting the dependency graph for learning I2C communication yields the following learning sequence: "I2C Basic Concepts" → "I2C Communication Principles" → "STM32 Hardware I2C Interface" → "I2C Communication Program Implementation." This sequence ensures that students have mastered the necessary prerequisite knowledge before learning each knowledge point.

[0049] Next, the number N of surrounding knowledge points that each knowledge gap node needs to include can be calculated based on the knowledge depth parameter in the adaptive teaching plan. The calculation method can adopt a preset segmented mapping relationship, for example: when the knowledge depth parameter is in the range of 0 to 0.5, set N=1; when the knowledge depth parameter is in the range of 0.5 to 1.0, set N=2, etc.; according to this mapping relationship, the number of surrounding knowledge points N is determined. From the cognitive model, the N surrounding knowledge points with the lowest familiarity with each gap node can be selected and added to the learning path. For example, for the knowledge gap node "I2C Communication Principles", for example, the three surrounding knowledge points with the lowest familiarity, "Serial Communication Basics", "Clock Synchronization Mechanism" and "Address Addressing Method", are selected as auxiliary learning content.

[0050] Based on the knowledge depth coefficient, the number of content levels M for each knowledge point can be determined. This number can be calculated using a linear mapping method. For example, if M = 4, this indicates that each knowledge point should contain four levels of content with increasing difficulty, including basic concepts, principle analysis, application methods, and problem-solving. This vertical learning structure enables students to understand each knowledge point from the shallow end to the deep end, gradually increasing their cognitive depth.

[0051] Finally, the learning sequence generated by topological sorting can be used as the backbone, combined with each node's N surrounding knowledge points to form a horizontal expansion network, and each node's M content levels to form a vertical depth structure, forming a complete customized learning sequence. This three-dimensional learning structure ensures a logical order in knowledge learning while providing both breadth and depth. This customized learning sequence, constructed in this way, fully considers the dependencies between knowledge points, students' cognitive characteristics, and the integrity of their knowledge structure. It provides students with a learning path that conforms to learning patterns and adapts to their individual characteristics, effectively improving learning efficiency and effectiveness.

[0052] The learning content generation module is used to generate learning content based on customized learning sequences, learning material combination strategies and explanation expression strategies.

[0053] Generating learning content based on customized learning sequences, learning material combination strategies, and explanation and expression strategies includes: Generate content for each level of each node in the customized learning sequence according to the content control strategy; Content control strategies include: controlling the proportion of various types of learning materials in the generated content based on the learning material combination strategy, and controlling the number of professional terms and corresponding concept explanations in the generated content based on the explanation expression strategy.

[0054] In this embodiment, the learning content generation module generates learning content according to the customized learning sequence, learning material combination strategy and explanation expression strategy to provide students with personalized learning materials.

[0055] First, content can be generated for each node and level in the customized learning sequence according to the content control strategy. The content control strategy mainly includes two aspects: the control of learning material combination and the control of explanation expression, ensuring that the generated content is consistent with students' learning preferences and cognitive level.

[0056] Based on the learning material combination strategy, the proportion of each type of learning material in the generated content can be controlled. For example, for the aforementioned embedded systems learner, the learning material combination strategy is 27% video tutorials, 29% interactive experiments, 23% text tutorials, and 21% diagrams. For the "Principle Analysis" level of the knowledge point "I2C Communication Principles," the learning content can be organized according to the following ratio: providing approximately 27% video explanations, such as a video analyzing I2C communication timing waveforms; designing approximately 29% interactive experiments, such as a simulator that visualizes the I2C communication process; writing approximately 23% text tutorials detailing the I2C communication principles; and configuring approximately 21% diagrams, such as I2C bus timing diagrams and state transition diagrams. This material combination fully considers students' learning preferences and helps improve learning efficiency and experience.

[0057] Based on the explanation strategy, you can control the amount of specialized terms and corresponding concept explanations in the generated content. For example, if the explanation strategy stipulates a maximum of 30% for specialized terms and a two-level depth for concept explanation, then when generating learning content for "I2C Communication Principles," you can limit specialized terms (such as "clock synchronization," "addressing," and "response signal") to no more than 30% of the total vocabulary. At the same time, these specialized terms should be explained at a maximum of two levels of depth: first, the specialized terms should be explained using basic concepts, and then the basic concepts should be explained as necessary, but no further explanation of the underlying concepts should be given.

[0058] For example, when explaining the "clock synchronization mechanism in I2C communication," one could state: "I2C communication uses a clock synchronization mechanism to ensure reliable data transmission. Clock synchronization refers to the process by which the master and slave devices coordinate the timing of data transmission using a shared clock signal (SCL line). During this process, the slave device can extend the clock cycle by pulling the SCL line low, achieving speed matching. This technique is called clock stretching." In this explanation, "clock synchronization" is a technical term, explained at the first level as "the process of coordinating data transmission timing." "Clock stretching" is another technical term, explained at the second level as "pulling the SCL line low to extend the clock cycle and achieve speed matching." This expression ensures both professional accuracy and comprehensibility.

[0059] In the actual generation process, different generation strategies can be adopted for different types of learning materials. For video tutorials, scripts can be designed to control the explanation rhythm and the frequency of professional terminology. For interactive experiments, multi-level operational guidance can be designed, ranging from basic operations to advanced applications. For text tutorials, the proportion of professional terminology and the depth of explanation can be precisely controlled. For charts and graphs, different levels of annotations can be added. The learning content generated in this way not only meets students' preferences for different learning materials, but also adapts to their cognitive levels, providing a personalized learning experience for each level of each knowledge point. At the same time, this generation method ensures a balance between professionalism and comprehensibility in the learning content, effectively supporting students' knowledge construction process.

[0060] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0061] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A multimodal artificial intelligence learning assistant based on an embedded platform, characterized in that: include: The data collection and processing module is used to receive learning questions and student learning data, and standardize the learning questions to obtain a standardized data set of questions; Cognitive model building module, which is used to build a learner cognitive model based on the standardized problem dataset and student learning data, and mark the knowledge gap nodes in the learner cognitive model; A teaching plan generation module is used to determine an adaptive teaching plan based on the structural characteristics of the learner's cognitive model. The adaptive teaching plan includes a knowledge depth parameter, a learning material combination strategy, and an explanation and expression strategy. A learning sequence building module is used to build customized learning sequences based on knowledge gap nodes and knowledge depth parameters; The learning content generation module is used to generate learning content based on customized learning sequences, learning material combination strategies and explanation expression strategies.

2. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 1, characterized in that: The standardization of the learning problem to obtain a standardized problem dataset includes: Convert the learning question into question text, clean the question text, remove extra spaces and special characters, and obtain the cleaned question text; Perform word segmentation on the cleaned question text to divide the question text into semantic units, and identify subject entities, skill keywords, and operation points from the semantic units; Classify subject entities, skill keywords, and operation points into professional fields, corresponding to specific training modules and knowledge points; Integrate the cleaned question text, semantic units, subject entities, skill keywords, operation points and professional field classification results to generate a standardized question dataset.

3. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 2, characterized in that: The construction of a learner cognitive model based on the standardized problem dataset and student learning data includes: Extract student learning data, including relevant theoretical course grades, historical practical training scores, learning behavior records, and skill mastery levels; The subject entities, skill keywords and operation points identified in the standardized problem dataset are used as nodes to build an initial cognitive model; Based on the professional field classification results, the field association relationship between nodes is established and the field association relationship is mapped into the edge in the initial cognitive model; By analyzing students' theoretical course scores and skill mastery, we calculate students' mastery of each knowledge node and add mastery attributes to the nodes in the initial cognitive model. Based on students' historical learning behaviors and practical training scores, the students' familiarity with the knowledge association paths is calculated, and familiarity weight values ​​are assigned to the edges in the initial cognitive model to obtain the learner cognitive model.

4. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 3, characterized in that: The calculation of students' mastery of each knowledge node through analysis of students' theoretical course scores and skill mastery includes: Map each node in the initial cognitive model to the corresponding theoretical course knowledge points and practical training skill points; Extract students' historical performance data in corresponding theoretical courses and standardize the performance data to obtain theoretical mastery scores; Extract the scoring data of students' relevant skill points in history training, and standardize the scoring data to obtain the skill mastery score; The mastery degree is obtained by weighted summing the theoretical mastery score and the skill mastery score according to the preset weights.

5. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 4, characterized in that: The calculation of the student's familiarity with the knowledge association path based on the student's historical learning behavior and practical training scores includes: Analyze students' historical learning behavior data, including learning material access records, practical training operation sequences, and problem-solving paths; For each node pair connected by an edge in the initial cognitive model, extract the student's behavior sequence involving the node pair in historical learning; Calculate the average operation time and accuracy of students when solving problems related to this node; Familiarity was obtained based on average operation time and accuracy.

6. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 5, characterized in that: The marking of the knowledge gap nodes in the learner's cognitive model includes: Identify all nodes in the learner's cognitive model whose mastery attribute value is below a preset threshold; Obtain the connection relationship of these low-mastery nodes in the cognitive model and construct a knowledge gap subgraph; Calculate the importance score of each node based on its degree; Sort the nodes in the knowledge gap subgraph according to their importance scores, and select the first preset number of nodes with the highest scores as target nodes; Analyze the pre-knowledge dependencies of the target node and identify the basic knowledge points that must be mastered first; Mark the target node and necessary prerequisite knowledge points as knowledge gap nodes.

7. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 1, characterized in that: Determining the adaptive teaching plan based on the structural characteristics of the learner's cognitive model includes: The knowledge coverage coefficient is calculated based on the ratio of the number of knowledge gap nodes to the total number of cognitive model nodes; the knowledge depth coefficient is calculated based on the average in-and-out degree and association strength of the knowledge gap nodes; The knowledge coverage breadth coefficient and the knowledge depth coefficient are combined to form the knowledge depth parameter; Calculate the average time students spend on each type of learning material and the completion rate, normalize the time and completion rate, and then take the weighted sum to obtain the preference score for each type of learning material; According to the preference scores of various learning materials, weight values ​​are allocated proportionally, and the sum of the weights is 1 to form a learning material combination strategy; The complexity of content expression is calculated based on the average score of students' history grades and the technical complexity of knowledge gap nodes. Based on the complexity of content expression, the maximum proportion of professional terms in the total vocabulary and the depth of the level of explanation that must be given to the concepts are determined to form an explanation expression strategy; A complete adaptive teaching plan is formed by integrating the three dimensions of knowledge depth parameters, learning material combination strategies and explanation and expression strategies.

8. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 7, characterized in that: The construction of a customized learning sequence based on knowledge gap nodes and knowledge depth parameters includes: Based on the relationship between the previous learning requirements of the knowledge gap nodes, a learning dependency directed graph is constructed, where the direction of the edge indicates the direction from the previous knowledge to the subsequent knowledge; Apply the topological sorting algorithm to process the learning dependency directed graph and generate a knowledge point learning sequence that satisfies the principle of prioritizing learning of prior knowledge. Based on the knowledge depth parameter in the adaptive teaching plan, the number of surrounding knowledge points N that each knowledge gap node needs to include is calculated. The N surrounding knowledge points with the lowest familiarity with each gap node are selected from the cognitive model and added to the learning path as auxiliary learning content. Determine the number of content levels M for each knowledge point based on the knowledge depth coefficient, and plan M levels of content with increasing difficulty for each knowledge point, including basic concepts, principle analysis, application methods, problem solving, and advanced expansion, to form a vertical learning structure for knowledge points; The learning sequence generated by topological sorting is used as the backbone, combined with the N surrounding knowledge points of each node to form a horizontal expansion network, and the M content levels of each node to form a vertical depth structure, forming a complete customized learning sequence.

9. The multimodal artificial intelligence learning assistant based on an embedded platform according to claim 8, characterized in that: The generation of learning content based on customized learning sequence, learning material combination strategy and explanation expression strategy includes: Generate content for each level of each node in the customized learning sequence according to the content control strategy; The content control strategy includes: controlling the proportion of various types of learning materials in the generated content according to the learning material combination strategy, and controlling the number of professional terms and corresponding concept explanations in the generated content according to the explanation expression strategy.

Images