Personalized learning path recommendation system based on artificial intelligence

Abstract

The invention belongs to the technical field of artificial intelligence learning, and particularly relates to a personalized learning path recommendation system based on artificial intelligence, and the system comprises a learner state sensing module which is used for collecting and processing the multi-modal data of a learner in real time, and constructing a dynamic learner state model; the knowledge graph construction module is used for constructing a knowledge graph containing multi-modal learning resources; and the input end of the learning path generation engine is connected with the learner state sensing module and the knowledge graph construction module, and the learning path generation engine generates a personalized learning path sequence through a path planning algorithm based on the learner state and the knowledge graph structure. The scheme is completed by tracking and analyzing cognitive and emotional states of a learner in real time, the learner state sensing module continuously collects answer records, learning behaviors, facial expressions, voices and other multi-mode data, and the mastering probability of the learner on each knowledge point is accurately calculated by applying an item reaction theory.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence learning technology, specifically a personalized learning path recommendation system based on artificial intelligence. Background Technology

[0002] The core of interdisciplinary artificial intelligence lies in the integration of multiple disciplines. Traditional AI research often focuses on the field of computer science, but with the development of technology, AI has gradually penetrated into multiple industries such as medicine, education, finance, manufacturing, and transportation. This interdisciplinary application has not only broadened the scope of AI applications, but also promoted mutual learning and common development among various disciplines.

[0003] With the rapid development of information technology, online education platforms and intelligent learning tools have gradually become widespread. Traditional personalized learning systems mainly rely on users' historical behavioral data, such as answer accuracy, learning time, time spent on knowledge points, and preset learning goals, to plan learning sequences through algorithmic models. These systems have, to some extent, realized the initial concept of individualized instruction. According to the authorization announcement number CN120338082A, an intelligent personalized learning path recommendation system is disclosed, which relates to the field of learning system technology. It includes a multi-source data acquisition module for real-time synchronization of students' behavioral data in cross-platform learning scenarios; a cognitive feature analysis engine including a style recognition sub-engine and a demand prediction sub-engine to predict the intensity of students' potential learning needs for unmastered knowledge points and generate a priority list containing knowledge gaps; a personalized path generator to generate three-dimensional path planning; and automatically configures the proportion of guided questioning, example demonstrations, and independent exploration according to the difficulty of knowledge; a dynamic adaptive adjustment unit: based on a reinforcement learning framework, with students' real-time performance as the state space and path adjustment actions as the decision space, the reward function is designed to include short-term progress speed and long-term ability growth potential; and a dual-loop feedback mechanism is implemented. This invention breaks through the dimensional limitations of traditional learning analytics, overcomes the rigidity of static course templates, and bridges the semantic gap in subject cognition.

[0004] Existing personalized learning technologies still have significant shortcomings in terms of depth, breadth, and humanization, specifically in the following aspects:

[0005] 1. Most existing systems treat learners as rational, constant data processing terminals, completely ignoring the crucial emotional factors in the learning process. Learning is essentially a psychological process full of emotional fluctuations, and emotions such as confusion, fatigue, discouragement, or excitement will directly affect the efficiency of cognitive absorption.

[0006] 2. Although the current platform has accumulated a massive amount of multimedia resources such as text, video, and audio, they usually exist in the form of isolated and static resource libraries. The recommendation system often makes recommendations based on the tags or popularity of resources, and fails to deeply understand the complementary advantages of different modal resources in explaining the same knowledge point.

[0007] 3. Most personalized learning systems are designed with an emphasis on human-computer interaction, constructing learning as a closed and independent individual behavior. This ignores the crucial role of collaborative learning and social interaction in the knowledge internalization process. Summary of the Invention

[0008] The purpose of this invention is to provide an artificial intelligence-based personalized learning path recommendation system to solve the problems mentioned in the background art.

[0009] To address the above problems, this invention provides a technical solution: a personalized learning path recommendation system based on artificial intelligence, the personalized learning path recommendation system comprising:

[0010] The learner state awareness module is used to collect and process learners' multimodal data in real time and build a dynamic learner state model.

[0011] The knowledge graph construction module is used to build a knowledge graph that contains multimodal learning resources;

[0012] The learning path generation engine is connected to the learner state awareness module and the knowledge graph construction module at its input end. Based on the learner state and the knowledge graph structure, it generates personalized learning path sequences through a path planning algorithm.

[0013] The human-computer collaborative learning interface module is used to present recommended learning paths and support interaction between learners, the system, and other learners.

[0014] Preferably, the learner state awareness module includes:

[0015] The cognitive state assessment unit is used to assess learners' mastery of knowledge points based on their answer records, learning time, and content interaction behavior.

[0016] The cognitive state assessment unit uses an algorithm based on item response theory to quantify the learner's ability score. and the probability of mastering knowledge point k The calculation formula is:

[0017]

[0018] Where a is the discrimination index, b is the difficulty, and c is the guessing parameter. This represents the learner's cognitive ability score.

[0019] The emotion state recognition unit is used to identify the learner's current emotion state by analyzing the learner's facial expressions, tone of voice, and text interaction content. The emotion state includes at least confusion, fatigue, discouragement, and excitement.

[0020] Preferably, the emotion state recognition unit employs a multimodal fusion model based on an attention mechanism, fusing visual, audio, and text features; given visual feature V, audio feature A, and text feature T, the fused feature vector F is represented as:

[0021]

[0022] in, These are the weights of each modal feature calculated by the attention network. .

[0023] Preferably, the knowledge graph constructed by the knowledge graph construction module is a heterogeneous graph, whose nodes include:

[0024] Concept nodes represent discrete knowledge points;

[0025] Resource nodes represent multimodal learning resources that explain knowledge points, including text, video, audio, and interactive simulations;

[0026] The edges between the nodes include pre-revision relations, partial relations, and interpreted relations.

[0027] Preferably, the knowledge graph construction module further includes a resource complementarity evaluation unit, used to quantify the complementary advantages of two resources of different modalities in explaining the same knowledge point. and Their complementarity score Through the following methods: Specifically:

[0028] .

[0029] Preferably, the core of the learning path generation engine is a path planning agent based on deep reinforcement learning;

[0030] The state of the intelligent agent It consists of the learner's current state and the knowledge graph nodes already visited, and actions. The next target knowledge point to learn is the reward function. Designed as follows:

[0031] in, It is the weighting coefficient.

[0032] Preferably, the cognitive load in the reward function is specifically:

[0033]

[0034] Among them, the intrinsic workload is determined by the inherent complexity of the knowledge points, the additional workload is determined by the presentation method and information density of the learning resources, and the learner's state is determined by the learner's current emotional state and cognitive free capacity.

[0035] Preferably, the learning path generation engine employs a multi-objective optimization algorithm when generating paths, simultaneously optimizing learning efficiency, knowledge consolidation, and learning motivation; its objective function can be expressed as:

[0036]

[0037] in, To predict the knowledge points The degree of mastery, It is a knowledge point to learn and its corresponding resources The estimated time, It refers to the coherence between consecutive knowledge points;

[0038] It is a hyperparameter that balances different objectives.

[0039] Preferably, the human-machine collaborative learning interface module includes a social interaction integration unit, used for:

[0040] In the recommended learning path, identify the peer learners who are learning synchronously for specific knowledge points;

[0041] Provides entry points to discussion forums based on knowledge points, and highlights discussion topics that share similar concerns with the current learner;

[0042] Based on the progress of the learning path, the system intelligently recommends the formation of temporary, goal-oriented collaborative learning groups.

[0043] Preferably, the personalized learning path recommendation system further includes a metacognitive feedback loop module, used for:

[0044] Regularly demonstrate the learner's learning path, status evolution, and the logic of the system's recommendations, and collect learner feedback;

[0045] The reward function weights in the learning path generation engine or the parameters in the learner state awareness model are dynamically adjusted based on the learner's direct and indirect feedback.

[0046] The beneficial effects of this invention are:

[0047] 1. This solution achieves this by tracking and analyzing learners' cognitive and emotional states in real time. The learner state perception module continuously collects multimodal data such as answer records, learning behaviors, facial expressions, and voice. It uses item response theory to accurately calculate the probability of mastering each knowledge point. At the same time, it integrates multiple features through attention mechanisms to identify emotional states such as confusion or excitement. These dynamically updated state models are combined with a graph containing rich learning resources and knowledge relationships. The learning path generation engine uses this information and deep reinforcement learning algorithms to plan a unique learning sequence for each learner. This sequence not only targets knowledge gaps but also actively avoids learning obstacles caused by emotional fatigue or cognitive overload, thereby ensuring that each learning suggestion is in line with the learner's current actual ability and psychological state.

[0048] 2. This solution significantly improves overall learning efficiency and sustained motivation. Its core lies in the multi-objective optimization strategy employed by the learning path generation engine. This strategy simultaneously balances learning efficiency, knowledge consolidation, and the coherence of learning content, aiming to help learners achieve a more solid level of mastery in a shorter time. Furthermore, the system quantifies the cognitive load during the learning process by assessing the inherent complexity and presentation of learning resources, and proactively selects resource combinations that optimize load and reduce redundant content when recommending paths. Simultaneously, the human-computer collaborative learning interface module embeds social interaction functions, enabling learners to identify learning partners, highlight relevant discussions, and intelligently form temporary learning groups. This combination of personalized paths and social support effectively alleviates the discouragement that may arise from independent learning, maintaining learners' enthusiasm and engagement through peer support and knowledge sharing.

[0049] 3. This solution periodically and transparently displays the learning path and the internal logic of the system's recommendations to learners, and proactively collects learners' subjective feedback. This direct feedback, along with indirect behavioral data generated during the interaction process, is used to dynamically adjust the parameters of the system's core algorithm. For example, the system can recalibrate the weights of various indicators in the reward function of the path generation engine or fine-tune the evaluation parameters in the state-aware model based on learners' long-term experience. This makes the system no longer a static tool, but an adaptive system that can grow with learners and continuously improve the recommendation strategy to better suit user habits and needs, thereby achieving continuous improvement in service quality. Attached Figure Description

[0050] For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0051] Figure 1 This is a flowchart of the personalized learning path recommendation system of the present invention. Detailed Implementation

[0052] Example 1

[0053] like Figure 1 As shown, this specific embodiment adopts the following technical solution: a personalized learning path recommendation system based on artificial intelligence, the personalized learning path recommendation system comprising:

[0054] The learner state awareness module is used to collect and process learners' multimodal data in real time and build a dynamic learner state model.

[0055] The knowledge graph construction module is used to build a knowledge graph that contains multimodal learning resources;

[0056] The learning path generation engine is connected to the learner state awareness module and the knowledge graph construction module at its input end. Based on the learner state and the knowledge graph structure, it generates personalized learning path sequences through path planning algorithms.

[0057] The human-computer collaborative learning interface module is used to present recommended learning paths and support interaction between learners, the system, and other learners.

[0058] The learner state awareness module includes:

[0059] The cognitive state assessment unit is used to assess learners' mastery of knowledge points based on their answer records, learning time, and content interaction behavior.

[0060] The cognitive state assessment unit uses an algorithm based on item response theory to quantify learners' ability scores. and the probability of mastering knowledge point k The calculation formula is:

[0061]

[0062] Where a is the discrimination index, b is the difficulty, and c is the guessing parameter. This represents the learner's cognitive ability score.

[0063] The emotion state recognition unit is used to identify the learner's current emotion state by analyzing the learner's facial expressions, tone of voice, and text interaction content. The emotion state includes at least confusion, fatigue, discouragement, and excitement.

[0064] The emotion state recognition unit employs a multimodal fusion model based on an attention mechanism, fusing visual, audio, and text features. Given visual feature V, audio feature A, and text feature T, the fused feature vector F is represented as:

[0065]

[0066] in, These are the weights of each modal feature calculated by the attention network. .

[0067] The knowledge graph constructed by the knowledge graph construction module is a heterogeneous graph, and its nodes include:

[0068] Concept nodes represent discrete knowledge points;

[0069] Resource nodes represent multimodal learning resources that explain knowledge points, including text, video, audio, and interactive simulations;

[0070] The edges between nodes include pre-relation relations, partial relations, and interpreted relations.

[0071] The knowledge graph construction module also includes a resource complementarity evaluation unit, used to quantify the complementary advantages of two different modalities in explaining the same knowledge point. and Their complementarity score Through the following methods: Specifically:

[0072] .

[0073] The core of the learning path generation engine is a path planning agent based on deep reinforcement learning;

[0074] The state of the agent It consists of the learner's current state and the knowledge graph nodes already visited, and actions. The next target knowledge point to learn is the reward function. Designed as follows:

[0075] in, It is the weighting coefficient.

[0076] Specifically, the cognitive load in the reward function is as follows:

[0077]

[0078] Among them, the intrinsic workload is determined by the inherent complexity of the knowledge points, the additional workload is determined by the presentation method and information density of the learning resources, and the learner's state is determined by the learner's current emotional state and cognitive free capacity.

[0079] The learning path generation engine employs a multi-objective optimization algorithm when generating paths, simultaneously optimizing learning efficiency, knowledge consolidation, and learning motivation; its objective function can be expressed as:

[0080]

[0081] in, To predict the knowledge points The degree of mastery, It is a knowledge point to learn and its corresponding resources The estimated time, It refers to the coherence between consecutive knowledge points;

[0082] It is a hyperparameter that balances different objectives.

[0083] The human-machine collaborative learning interface module includes a social interaction integration unit, used for:

[0084] In the recommended learning path, identify the peer learners who are learning synchronously for specific knowledge points;

[0085] Provides entry points to discussion forums based on knowledge points, and highlights discussion topics that share similar concerns with the current learner;

[0086] Based on the progress of the learning path, the system intelligently recommends the formation of temporary, goal-oriented collaborative learning groups.

[0087] The personalized learning path recommendation system also includes a metacognitive feedback loop module, used for:

[0088] Regularly demonstrate the learner's learning path, status evolution, and the logic of the system's recommendations, and collect learner feedback;

[0089] Based on learners' direct and indirect feedback, the reward function weights in the learning path generation engine or the parameters in the learner state-aware model are dynamically adjusted.

[0090] Example 2

[0091] A personalized learning path recommendation method based on artificial intelligence is as follows:

[0092] S1. Real-time data collection: The learner status awareness module starts working and collects multimodal data of learners in real time, including answer records, learning time, content interaction behavior, as well as facial expressions, voice tone and text content;

[0093] S2. State Modeling and Evaluation:

[0094] Cognitive Status Assessment: The cognitive status assessment unit utilizes the item response theory model to quantify the probability of learners mastering each knowledge point based on collected behavioral data such as answering questions. and cognitive ability score ;

[0095] Emotional state recognition: The emotional state recognition unit uses an attention-based multimodal fusion model to weightedly fuse visual, audio, and text features to identify the learner's current emotional state, such as confusion or fatigue.

[0096] S3. Constructing a knowledge graph: The knowledge graph construction module builds and maintains a heterogeneous graph containing concept nodes and resource nodes, where the resource complementarity evaluation unit calculates the complementarity score of different modal resources when explaining the same knowledge point;

[0097] S4. Generate Path Action: The learning path generation engine receives the learner state model from the learner state awareness module and the knowledge graph from the knowledge graph construction module. Based on the current state, the deep reinforcement learning agent will select an action, namely, recommend the next most suitable learning target knowledge point. ;

[0098] S5. Calculate Instant Rewards: While generating path actions, the system will calculate rewards based on the pre-designed reward function. The immediate reward for the action is calculated, which takes into account multiple factors such as improved mastery, learner engagement, estimated cognitive load, and content redundancy.

[0099] S6. Optimize the learning sequence: The learning path generation engine uses a multi-objective optimization algorithm, which not only considers the reward of a single step, but also plans a complete and continuous learning path sequence to simultaneously optimize long-term goals such as learning efficiency, knowledge consolidation and learning motivation.

[0100] S7. Presentation and Interaction: The human-computer collaborative learning interface module presents the learning path sequence generated by the system to the learner. At the same time, the social interaction integration unit will mark the knowledge points on the path with peer learners, provide the discussion area entrance, and may recommend the formation of temporary collaborative learning groups.

[0101] S8. Learner Execution and Feedback: Learners follow the recommended learning path and interact with learning resources, the system, and other learners through the human-computer interface. During this process, the system continuously executes steps S1 and S2 to update the learner's status.

[0102] S9. Collect metacognitive feedback: The metacognitive feedback loop module regularly shows learners their learning path and system recommendation logic, and actively collects learners' direct evaluations, such as "too easy" or "didn't understand", as well as indirect behavioral feedback.

[0103] S10. System Dynamic Optimization: Based on the new data collected by the metacognitive feedback loop, the system dynamically adjusts the relevant models and parameters.

[0104] Example 3

[0105] Application scenario: Adaptive learning for the "Microeconomics" course in higher education;

[0106] Background: A university introduced this personalized learning path recommendation system into its online Microeconomics course, aiming to help hundreds of students from diverse backgrounds overcome learning difficulties and improve pass rates and deep learning outcomes.

[0107] M1. System Operation Process:

[0108] Initial state perception and knowledge graph construction:

[0109] The knowledge graph construction module has built a detailed knowledge graph for "Microeconomics". Concept nodes include "law of supply and demand", "elasticity", "consumer surplus" and "market failure", while resource nodes include professor's lecture videos, e-textbook chapters, interactive supply and demand curve simulators, and related academic podcast audio.

[0110] Student Xiao Li begins learning, and the learner state awareness module starts working:

[0111] The cognitive assessment unit found through Xiao Li's pre-test that he had a high accuracy rate in answering questions on the basic concept of "the law of supply and demand," and his ability score was high. High probability, master the probability However, regarding the knowledge point of "elasticity," the answer record showed repeated errors, and the system calculated the probability of mastery. ;

[0112] The emotional state recognition unit detected that Xiao Li frequently frowned and sighed when learning about "elasticity" through his camera and microphone. At the same time, he wrote in the forum question for this chapter: "I can never distinguish between the price elasticity of demand and the income elasticity, it's so annoying..." After comprehensive analysis by the multimodal fusion model, his emotional state was identified as "confused" and "disheartened".

[0113] M2, Personalized Route Generation:

[0114] The learning path generation engine receives Xiao Li's state: cognitively, he has weaknesses in "flexible" knowledge points; emotionally, he is "confused" and "disheartened." The engine uses this state as its basis. Start planning the next steps. ;

[0115] reward function calculate:

[0116] If we directly recommend the more difficult "consumer surplus", the cognitive load will be very high, and the expected improvement in mastery will be negative, resulting in a very low score in the reward function.

[0117] If the system recommends relearning "elasticity," it will utilize the resource complementarity assessment unit. It finds that Xiao Li previously primarily viewed text-based materials, and now recommends an interactive simulator and a short animated video. These two resource modalities are significantly different but semantically similar, resulting in a high complementarity score. High, which can reduce additional load and increase engagement;

[0118] At the same time, the system balances learning efficiency, knowledge consolidation, and learning motivation through multi-objective optimization;

[0119] Decision: The final path generated by the engine is not simply to have Xiao Li reread the textbook, but rather: [Watch the "elasticity" animation video] → [Explore using the interactive simulator] → [Complete a small quiz on the concept of "elasticity"];

[0120] M3, Human-Machine Collaboration and Metacognitive Feedback:

[0121] When recommending the above paths, the social interaction integration unit played a role in the human-machine collaborative learning interface module:

[0122] Next to the knowledge point about "elasticity", it is noted that "15 students are currently learning this knowledge point";

[0123] A discussion topic was highlighted: "What are some goods with negative income elasticity of demand? Please provide examples." Several students with similar questions to Xiao Li were discussing this topic.

[0124] The system intelligently recommended and formed a temporary study group for two days to "tackle the concept of elasticity" for Xiao Li and two other students who also spent a lot of time on the knowledge point of "elasticity".

[0125] Xiao Li completed the recommended path through the API and gained inspiration during the group discussion. After a specialized test, his mastery rate increased to [percentage missing]. The emotion recognition also shifted to "calm" and "slight sense of accomplishment";

[0126] The metacognitive feedback loop module then showed Xiao Li his learning path over the past week and explained, "We found that you are having difficulty with 'elasticity,' so we recommended visualization resources and group collaboration to reduce cognitive load and deepen understanding through peer learning." The system invited Xiao Li to rate the recommendation, and Xiao Li gave it a five-star rating. This feedback will be used to fine-tune the weights of the reward function in the engine.

[0127] The innovation of this invention lies in the following: In the learner state perception module, a dynamic learner state model is constructed by real-time collection and processing of multimodal data. The cognitive state assessment unit quantifies the learner's ability value and the probability of mastering knowledge points based on the item response theory algorithm. This algorithm constructs a nonlinear function relationship through discrimination parameters, difficulty parameters, and guessing parameters, which can accurately reflect the learner's response characteristics to knowledge points of different difficulties. The emotional state recognition unit adopts a multimodal fusion model based on the attention mechanism. Through adaptive weighted fusion of visual features, audio features, and text features, the attention network dynamically calculates the contribution weight of each modality feature, ensuring accurate identification of complex emotional states, including confusion, fatigue, discouragement, and excitement, based on facial expression analysis, speech tone recognition, and text content understanding. This multidimensional state perception mechanism provides the system with a comprehensive and dynamically updated learner profile, enabling subsequent path planning to be accurately adapted based on both cognitive and emotional dimensions.

[0128] The knowledge graph construction module establishes a heterogeneous graph network containing concept nodes and resource nodes. Concept nodes represent discrete knowledge point units, while resource nodes are associated with multimodal learning materials such as text, video, audio, and interactive simulations. Nodes form a structured knowledge system through prerequisite relations, partial relations, and explained relations. The resource complementarity evaluation unit of this module quantifies the synergistic value of different modal resources in explaining the same knowledge point by calculating the product of the semantic similarity of the resource content embedding vector and the diversity measure of resource modality differences. This design ensures the core semantic consistency of the learning content and enhances the diversity of knowledge presentation through modal combination. The learning path generation engine based on deep reinforcement learning combines this knowledge graph with the learner state model. Its agent achieves a dynamic balance between knowledge exploration and cognitive load during the path planning process by using a multi-factor reward function that includes mastery improvement, input incentive, cognitive load regulation, and content redundancy control through state space modeling and action space selection.

[0129] The system employs a multi-objective optimization framework to coordinate key indicators such as learning efficiency, knowledge consolidation, and learning motivation. Its objective function integrates optimization goals—including expected mastery improvement, minimization of learning time, and maximization of coherence between knowledge points—through hyperparameter weighting, forming a comprehensive path evaluation standard. The human-computer collaborative learning interface module enhances the collaborative nature of the learning process through social interaction integration units, real-time labeling of synchronous learning partners, intelligent matching of discussion topics, and formation of temporary collaborative groups, organically combining personalized and social learning. The metacognitive feedback loop module visualizes the learning path and system decision-making logic, collecting direct and indirect feedback from learners to form a dynamic calibration mechanism for the reward function weights of the learning path generation engine and the parameters of the learner state perception model. This two-way feedback structure ensures that the system continuously optimizes its recommendation strategy as the learning process progresses, achieving a truly adaptive learning experience.

[0130] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the present invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope. All such changes and modifications fall within the scope of the present invention as claimed, which is defined by the appended claims and their equivalents.

Claims

1. A personalized learning path recommendation system based on artificial intelligence, characterized in that, The personalized learning path recommendation system includes: The learner state awareness module is used to collect and process learners' multimodal data in real time and build a dynamic learner state model. The knowledge graph construction module is used to build a knowledge graph that contains multimodal learning resources; The learning path generation engine is connected to the learner state awareness module and the knowledge graph construction module at its input end. Based on the learner state and the knowledge graph structure, it generates personalized learning path sequences through a path planning algorithm. The human-computer collaborative learning interface module is used to present recommended learning paths and support interaction between learners, the system, and other learners.

2. The personalized learning path recommendation system based on artificial intelligence according to claim 1, characterized in that: The learner state awareness module includes: The cognitive state assessment unit is used to assess learners' mastery of knowledge points based on their answer records, learning time, and content interaction behavior. The cognitive state assessment unit uses an algorithm based on item response theory to quantify the learner's ability score. and the probability of mastering knowledge point k The calculation formula is: ； Where a is the discrimination index, b is the difficulty, and c is the guessing parameter. This represents the learner's cognitive ability score. The emotion state recognition unit is used to identify the learner's current emotion state by analyzing the learner's facial expressions, tone of voice, and text interaction content. The emotion state includes at least confusion, fatigue, discouragement, and excitement.

3. The personalized learning path recommendation system based on artificial intelligence according to claim 2, characterized in that: The emotion state recognition unit employs a multimodal fusion model based on an attention mechanism, fusing visual, audio, and text features. Given visual feature V, audio feature A, and text feature T, the fused feature vector F is represented as: ； in, These are the weights of each modal feature calculated by the attention network. .

4. The personalized learning path recommendation system based on artificial intelligence according to claim 3 is characterized in that: The knowledge graph constructed by the knowledge graph construction module is a heterogeneous graph, whose nodes include: Concept nodes represent discrete knowledge points; Resource nodes represent multimodal learning resources that explain knowledge points, including text, video, audio, and interactive simulations; The edges between the nodes include pre-revision relations, partial relations, and interpreted relations.

5. The personalized learning path recommendation system based on artificial intelligence according to claim 4, characterized in that: The knowledge graph construction module also includes a resource complementarity evaluation unit, used to quantify the complementary advantages of two different modalities in explaining the same knowledge point. and Their complementarity score Through the following methods: Specifically: 。 6. The personalized learning path recommendation system based on artificial intelligence according to claim 5, characterized in that: The core of the learning path generation engine is a path planning agent based on deep reinforcement learning; The state of the intelligent agent It consists of the learner's current state and the knowledge graph nodes already visited, and actions. The next target knowledge point to learn is the reward function. Designed as follows: in, It is the weighting coefficient.

7. The personalized learning path recommendation system based on artificial intelligence according to claim 6, characterized in that: The cognitive load in the reward function is specifically: ； Among them, the intrinsic workload is determined by the inherent complexity of the knowledge points, the additional workload is determined by the presentation method and information density of the learning resources, and the learner's state is determined by the learner's current emotional state and cognitive free capacity.

8. The personalized learning path recommendation system based on artificial intelligence according to claim 7, characterized in that: The learning path generation engine employs a multi-objective optimization algorithm when generating paths, simultaneously optimizing learning efficiency, knowledge consolidation, and learning motivation. Its objective function can be expressed as: ； in, To predict the knowledge points The degree of mastery, It is a knowledge point to learn and its corresponding resources The estimated time, It refers to the coherence between consecutive knowledge points; It is a hyperparameter that balances different objectives.

9. The personalized learning path recommendation system based on artificial intelligence according to claim 8, characterized in that: The human-machine collaborative learning interface module includes a social interaction integration unit for: In the recommended learning path, identify the peer learners who are learning synchronously for specific knowledge points; Provides entry points to discussion forums based on knowledge points, and highlights discussion topics that share similar concerns with the current learner; Based on the progress of the learning path, the system intelligently recommends the formation of temporary, goal-oriented collaborative learning groups.

10. The personalized learning path recommendation system based on artificial intelligence according to claim 9, characterized in that: The personalized learning path recommendation system also includes a metacognitive feedback loop module, used for: Regularly demonstrate the learner's learning path, status evolution, and the logic of the system's recommendations to the learner, and collect learner feedback; The reward function weights in the learning path generation engine or the parameters in the learner state awareness model are dynamically adjusted based on the learner's direct and indirect feedback.

Images