Education course recommendation system based on big data

Abstract

The invention discloses an education course recommendation system based on big data, and particularly relates to the crossing field of education information and artificial intelligence, and the system comprises a data perception and fusion module which collects education data, generates behavior atoms, and forms a native education data space; the individual-group collaborative evolution module is used for reading behavior atoms from a data space, constructing an individual cognitive coordinate system and dynamically drawing a group knowledge potential energy diagram; the course resource labeling and organizing module is used for labeling dynamic cognitive load indexes and knowledge bridging values for course resources and embedding knowledge potential energy diagrams; and the self-adaptive guide path generation module is used for performing multi-step prospective exploration by taking the current cognitive coordinate of the user as a starting point and combining the knowledge potential energy diagram, the cognitive load and the bridging value, and generating a course sequence with the lowest cognitive energy consumption as a recommendation result. According to the method, planning from shallow resources to deep cognitive paths is realized, and course recommendation accuracy and learning efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of educational information and artificial intelligence, and more specifically, to an educational course recommendation system based on big data. Background Technology

[0002] With the deep integration of information technology and education, online learning platforms have accumulated massive amounts of user behavior data and course resources. How to extract value from this big data and provide learners with accurate and personalized course recommendations to improve learning efficiency and experience has become a key challenge in the field of educational technology. Existing recommendation technologies have alleviated the problem of "information overload" to some extent, but still have limitations in terms of intelligence and depth.

[0003] Existing technical solutions are mainly based on collaborative filtering and content tag matching. The system usually analyzes users' historical course selection and rating records to find similar users or similar courses for recommendation; or it assigns static tags (such as subject and difficulty) to courses and users and performs keyword matching. The implementation process relies on explicit rating behavior and a predefined tag system, and the recommendation list is generated by calculating the similarity matrix between users or courses.

[0004] However, in practical use, it still has some shortcomings, such as heavy reliance on explicit feedback data (such as ratings), making it difficult to capture implicit behaviors that contain rich cognitive information, such as video dragging, dwell time, and note annotations, resulting in shallow user modeling; course resources usually use fixed and static difficulty labels, which cannot dynamically reflect the real workload and effect of users with different cognitive levels when learning the course; the recommendation logic is essentially a simple association or matching of resource items, which cannot construct a coherent learning path that conforms to the laws of cognitive development, making it difficult to achieve true "personalized instruction" and adaptive learning. This solution aims to fundamentally solve these shortcomings. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an education course recommendation system based on big data, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an education course recommendation system based on big data, including a data perception and fusion module: an embedded semantic disambiguation and context reconstruction unit, which collects native education data, performs cross-modal intent extraction and context anchoring, generates behavioral atoms with unified spatiotemporal stamps and semantic contexts and stores them in a distributed file system to form a native education data space; Individual-Group Co-evolution Module: Asynchronously reads behavioral atoms from the native education data space, performs individual learning path tracking and group knowledge topology mining, deconstructs the intensity of user input and cognitive depth in different knowledge units, constructs a personal cognitive coordinate system that reflects the dynamic knowledge state, analyzes the co-occurrence and transmission patterns between group behavioral atoms, and dynamically draws a group knowledge potential energy map. Course Resource Labeling and Organization Module: This module analyzes course resources in the native educational data space, examines the mapping relationship between course content and the group knowledge potential energy map, and analyzes the effect feedback after consumption by users at different cognitive coordinate points. It labels the dynamic cognitive load index and knowledge bridging value, embedding them into the group knowledge potential energy map in the form of nodes. The adaptive guidance path generation module receives real-time requests from users and their corresponding personal cognitive coordinate system. It takes the user's current cognitive coordinates as the starting point, the group knowledge potential energy map as the navigation map, the course cognitive load index as the slope reference, and the course knowledge bridging value as the path selection basis. It conducts multi-step forward-looking path exploration, generates the lowest cognitive energy path from the current state to the target knowledge area, and outputs the course sequence mapped by the path as the final recommendation result.

[0007] The technical effects and advantages of this invention are as follows: 1. This invention constructs "behavioral atoms" that integrate multimodal data and cognitive auxiliary signals, and on this basis creates a dynamic "personal cognitive coordinate system" and "group knowledge potential energy map" to deeply analyze the user's knowledge status, cognitive load and group knowledge structure, thereby getting rid of the dependence on historical ratings or static tags and realizing accurate recommendations based on deep cognitive understanding; 2. This invention generates the optimal learning sequence from the current state to the target area for users by comprehensively considering the cognitive load (gradient) and knowledge bridging value (path efficiency) of the course. This ensures that the recommended learning path is not just a collection of related courses, but a personalized navigation scheme that conforms to cognitive laws and can efficiently achieve learning goals. 3. This invention achieves dynamic updates based on continuously flowing "behavioral atoms," enabling the annotation of user profiles, group knowledge graphs, and course resources. Simultaneously, the system has a dynamic path adjustment interface, which re-plans paths based on real-time user learning feedback. This allows the recommendation system to continuously evolve along with the user's cognitive state and the evolution of group knowledge, becoming a dynamic and adaptive learning guidance system. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a detailed schematic diagram of the data sensing and fusion module of the present invention; Figure 3A schematic diagram illustrating the construction details of the personal cognitive coordinate system of this invention; Figure 4 This is a schematic diagram illustrating the details of the course resource annotations in this invention; Figure 5 This is a schematic diagram showing the details of the adaptive guidance path generation of the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] As attached Figure 1 The system shown is an education course recommendation system based on big data, including a data perception and fusion module: an embedded semantic disambiguation and context reconstruction unit, which collects native education data, performs cross-modal intent extraction and context anchoring, generates behavioral atoms with unified spatiotemporal stamps and semantic contexts and stores them in a distributed file system to form a native education data space.

[0011] It should be specifically noted that the native educational data includes multimodal learning behavior data, context-related data, and cognitive auxiliary data; the semantic disambiguation is based on the knowledge graph and behavioral semantic dictionary in the education field, adopts a pre-trained model enhanced by attention mechanism, and integrates attention and emotion features in cognitive auxiliary data to perform semantic parsing of ambiguous behaviors; the context reconstructor binds different modal data to a unified spatiotemporal stamp through a spatiotemporal alignment algorithm, and adopts a dynamic allocation mechanism of context weights to achieve context anchoring of behaviors.

[0012] It should be further explained that the "three-dimensional native data matrix" is constructed, and the modal learning behavior data covers video modalities, including learners' facial micro-expressions, gestures, and eye movements during classroom / self-study, which are collected through lightweight cameras with a sampling rate of 15 frames per second; audio modalities, including classroom speech and group discussion recordings, including tone intensity in decibels and pause intervals; text modalities, including homework answers, notes, and forum messages; and operation modalities, including courseware click trajectories, video dragging progress, and time spent answering questions.

[0013] Context-related data includes physical context data such as light intensity, noise level, and device network status in the learning environment, collected through smart terminal sensors; temporal context data such as learning start / end time, duration of a single learning session, and weekly / monthly learning frequency distribution; and task context data such as the current learning task type and associated course units.

[0014] Cognitive auxiliary data acquires attention concentration data, emotional state data, and actively marked "question points" and "error-prone points" labels during the learning process by integrating a lightweight brain-computer interface (which only collects blood oxygenation signals from the prefrontal cortex and is non-invasive).

[0015] As attached Figure 2 As shown, an "edge-cloud collaborative acquisition architecture" is adopted. The edge is deployed on the learner's terminal (mobile phone / tablet / learning machine) to collect video, audio, operation and contextual data in real time. Lightweight compression algorithms are used (video uses H.265+ customized region of interest cropping, retaining only the face and operation area) to reduce transmission bandwidth. Cognitive auxiliary data is collected by a dedicated portable device and transmitted via Bluetooth 5.3 Low Energy. In the preprocessing stage, a density-based DBSCAN improved algorithm (adapted to the sparsity characteristics of educational data) is used to remove abnormal data (such as invalid frames caused by camera obstruction and noise data caused by device failure). The audio data is subjected to voice activity detection (VAD) and noise reduction, and the text data is subjected to word segmentation, stop word deletion and part-of-speech tagging.

[0016] The semantic disambiguation engine constructs a dual-reference system of "knowledge graph in the education field + behavioral semantic dictionary," and adopts an attention-enhanced BERT model (EduBERT) to map preprocessed multimodal data into a unified feature vector. Combined with attention and emotion features from cognitive auxiliary data, it performs semantic parsing of ambiguous behaviors. For example, for the behavior of "repeatedly clicking on a courseware page," if high attention concentration and an emotion label of "confusion" are detected, it is parsed as "cognitive doubt about this knowledge point"; if attention is scattered and the emotion label is "pleasure," it is parsed as "non-learning operation."

[0017] The context reconstruction engine is based on a spatiotemporal alignment algorithm, which binds different modal data to a unified spatiotemporal stamp, accurate to the millisecond. It uses an NTP+ local clock calibration mechanism to ensure synchronization, constructs a "behavior-context" association graph, and introduces a "dynamic allocation of context weights" mechanism to adjust the weights of each context factor according to the learning task type, so as to achieve accurate context anchoring.

[0018] The disambiguated semantic information is integrated with the anchored context to generate a standard format of "behavioral atoms". Each atom contains six core fields: "spatial stamp + user ID + behavior type + semantic tag + context vector + cognitive state label". A hybrid storage architecture of "distributed file system + blockchain" is adopted. The original data of the behavioral atoms is stored in the HDFS distributed file system to ensure high throughput and scalability. Key metadata (user ID, spatiotemporal stamp, semantic tag hash value) is written into the consortium blockchain and endorsed by schools, teachers and technology providers to ensure data sovereignty and immutability, forming a traceable and verifiable native educational data space.

[0019] Individual-Group Co-evolution Module: Asynchronously reads behavioral atoms from the native educational data space, performs individual learning path tracking and group knowledge topology mining, deconstructs the intensity of user investment and cognitive depth in different knowledge units, constructs a personal cognitive coordinate system reflecting the dynamic knowledge state, analyzes the co-occurrence and transmission patterns between group behavioral atoms, and dynamically draws a group knowledge potential energy map.

[0020] As attached Figure 3 As shown, it should be specifically explained that the specific construction process of the personal cognitive coordinate system includes using the knowledge unit topology as the coordinate axis, the horizontal axis as the logical sequence of knowledge units, and the vertical axis as the comprehensive value determined by the cognitive depth feature set. Through a dynamic weighted factor analysis algorithm, the individual's input intensity and cognitive depth in each knowledge unit are transformed into coordinate values.

[0021] The intensity of investment is calculated by combining the proportion of learning time, the number of repeated learning sessions, and the frequency of active interaction. A cognitive decay coefficient is introduced to compensate for the decay of historical data. The cognitive depth is quantified by mapping behavior type to cognitive target classification level and combining answer accuracy and question resolution efficiency indicators.

[0022] It should be further explained that, taking the behavioral atoms in the native education data space as the core data source, we focus on extracting three types of key data. Individual behavioral sequence data extracts the behavioral atom sequences of a single user in the continuous time dimension, and focuses on filtering behaviors directly related to knowledge units (such as courseware learning, answering exercises, editing notes, and marking questions), and associates them with corresponding cognitive state annotations (attention, emotion).

[0023] The knowledge unit association data is based on the curriculum standards to construct the knowledge unit topology (including the relationship between prerequisite knowledge, parallel knowledge, and extended knowledge), and associates the set of behavioral atoms corresponding to each knowledge unit.

[0024] Group behavior interaction data extraction records of behavioral interactions between users in a group (such as speaking and responding relationships in group discussions, peer review of assignments, and Q&A interactions) and co-occurrence data of behavioral atoms on the same knowledge unit (such as multiple users learning the same knowledge point and marking questions at the same time).

[0025] An incremental data reading mechanism is adopted, which uses a Kafka message queue to monitor the atomic updates of the native educational data space in real time. Asynchronous reading is performed by user ID in shards to avoid system lag caused by batch reading. A "cognitive depth feature set" is constructed, including "input intensity features" and "cognitive depth features." Input intensity features include the learning time percentage of knowledge units, the number of repeated learning sessions, and the interaction frequency (such as the number of times questions are asked). Cognitive depth features are based on the "Bloom cognitive goal classification method," mapping cognitive levels to behavioral types (e.g., "memory" corresponds to browsing courseware, "application" corresponds to answering exercises, and "creation" corresponds to completing extension tasks). Cognitive depth is quantified by indicators such as answer accuracy and question resolution efficiency. The specific calculation formula is as follows: , in, The cognitive depth score is 1-6, corresponding to the 6 levels of Bloom's taxonomy. For the first Weights of class behaviors For the first The cognitive level of class behavior mapping is based on a basic score. For the first The correct answer rate corresponding to each type of behavior For the first Efficiency in resolving questions related to class behaviors.

[0026] Using the topological structure of knowledge units as the coordinate axis, with the horizontal axis representing the logical sequence of knowledge units and the vertical axis representing the comprehensive value determined by the cognitive depth feature set, the "dynamic weighted factor analysis" algorithm is employed to transform an individual's input intensity and cognitive depth in each knowledge unit into coordinate values. The formula for calculating input intensity is as follows: , in, The input intensity value is (0-1). The learning time for this knowledge unit. The total learning time for all knowledge units. This represents the number of times the knowledge unit is repeatedly studied. This represents the maximum number of repetitions across all knowledge units. This represents the number of times questions were asked proactively for this knowledge unit. To determine the maximum number of questions asked across all knowledge units, a "cognitive decay coefficient" is introduced, calculated using the following formula: , in, The cognitive decay coefficient is (0-1). This is the attenuation coefficient (default 0.01). The final coordinate value represents the number of days since the last time this knowledge unit was learned. Construct a real-time updated personal cognitive coordinate system. For example, if a student's cognitive depth in the "quadratic equations" knowledge unit is "application level" (CD=4), and their input intensity is 0.8, and it has been 3 days since their last study (α≈0.97), then the corresponding coordinates are (knowledge unit ID: K102, cognitive depth: 3.88, input intensity: 0.78).

[0027] The algorithm combines association rule mining and community discovery to identify high-frequency associated knowledge unit combinations based on the co-occurrence data of group behavior atoms, forming a group knowledge association network. The Louvain algorithm is used to divide knowledge communities and identify the common knowledge strengths and weaknesses of the group.

[0028] "Knowledge potential" is defined as "mean depth of group cognition × percentage of people who have mastered the knowledge × knowledge relevance", and the calculation formula is: , in, This represents the potential energy of knowledge. This represents the average cognitive depth of the group within this knowledge unit. To determine the number of users for this knowledge unit, Total number of users For the first The knowledge unit and the first The covariance of each related knowledge unit (reflecting the strength of the association). The maximum covariance value, To determine the number of related knowledge units, a heatmap visualization method is used to draw a group knowledge potential energy map. The red area represents the high potential energy area (high knowledge mastery and strong relevance), and the blue area represents the low potential energy area (weak knowledge and low relevance). A "real-time update mechanism" is established to recalculate the potential energy value every 24 hours based on newly added behavioral atoms, dynamically reflecting the evolution status of group knowledge.

[0029] Course Resource Annotation and Organization Module: This module analyzes course resources in the native educational data space, examines the mapping relationship between course content and the group knowledge potential energy map, and analyzes the effect feedback after consumption by users at different cognitive coordinate points. It annotates the dynamic cognitive load index and knowledge bridging value, embedding them into the group knowledge potential energy map in the form of nodes.

[0030] It should be specifically noted that the drawing of the group knowledge potential energy map adopts a fusion algorithm of association rule mining and community discovery to identify high-frequency associated knowledge unit combinations and divide knowledge communities. The knowledge potential energy is defined as a function of the average group cognitive depth, the proportion of people who have mastered the knowledge, and the knowledge association degree, and then the potential energy value of each knowledge unit is calculated.

[0031] The cognitive load index is a dynamic indicator based on the inherent characteristics of resources and user feedback. It is calculated by comprehensively considering the number of resource knowledge points, the complexity of information presentation, the average learning time of users, the error rate, and the frequency of attention distraction.

[0032] The knowledge bridging value is an efficiency indicator of how well resources help users reach a target cognitive state from their current state of understanding. It is calculated by combining the matching degree between resources and the group's knowledge potential map with the efficiency of the improvement in the user's cognitive depth after using the resources.

[0033] It should be further explained that, focusing on the correlation and matching between course resources and data space, three types of core data are selected. The multi-dimensional data of course resources include basic resource attributes (type: courseware / exercises / videos / cases; format: PDF / MP4 / PPT; duration / length), content structure data (knowledge point coverage, knowledge difficulty level, teaching objectives), and presentation format data (whether it includes animated demonstrations, interactive exercises, and case analysis).

[0034] Resource consumption feedback data extracts behavioral feedback (such as changes in answer accuracy after learning the resource, preference for subsequent similar resources, and evaluation tags for the resource) and changes in cognitive state (such as the degree of improvement in attention concentration and the degree of relief of confusion) of users at different cognitive coordinates after using course resources.

[0035] The group knowledge potential energy correlation data obtains the potential energy value, weak area distribution, and knowledge correlation path of each knowledge unit in the group knowledge potential energy map, and associates it with the corresponding course resources.

[0036] As attached Figure 4 As shown, a "multimodal content parsing engine" is used to perform in-depth analysis of different types of resources: text resources (courseware, exercises) are extracted using the EduBERT model to extract information such as knowledge point tags, difficulty levels, and cognitive objectives; video resources are analyzed by combining speech recognition (extracting knowledge points), image recognition (extracting demonstration cases), and temporal analysis (extracting teaching rhythm) to analyze the content structure; interactive resources (virtual experiments, interactive exercises) are analyzed to analyze the interactive logic, feedback mechanisms, and knowledge point examination methods. The analysis results form a "resource-knowledge point" association matrix, marking the coverage strength of each knowledge point.

[0037] Breaking away from the traditional "static difficulty labeling" model, a load calculation model is constructed based on "resource characteristics + user feedback," with the following calculation formula: , in, Cognitive load index (1-10 points). For the number of resource knowledge points, For resource duration, Rate the complexity of the information presentation (1-5 points). The average learning time for users, The average error rate for users. The average frequency of user attention loss is calculated; the coefficients are optimized using regression analysis algorithms and dynamically updated based on changes in the user group; for example, the load index of the same exercise is 8 for users at the "basic" cognitive level and 4 for users at the "advanced" level.

[0038] The bridging value is defined as "the efficiency with which resources help users reach a target cognitive state from their current cognitive state." Combining two dimensions, the calculation formula is as follows: , in, Score the bridging value (1-5 points). The matching degree between resources and group knowledge potential map (1-5 points, weighted by the degree of coverage of weak areas and the degree of fit of related paths). To improve conversion efficiency, , This represents the average increase in user cognitive depth after using the resources. The average learning time was used; the analytic hierarchy process (AHP) was employed to determine the weights and calculate the bridging value score.

[0039] The labeled course resources are embedded as "nodes" into the collective knowledge potential graph. Resource nodes are associated with corresponding knowledge unit nodes. The node size maps to the bridging value score, and the node color maps to the cognitive load index, with green indicating low load, yellow indicating medium load, and orange indicating high load. A three-element association index of "resource-knowledge-user" is established to support multi-dimensional resource retrieval based on knowledge unit, cognitive load, and bridging value.

[0040] The adaptive guidance path generation module receives real-time requests from users and their corresponding personal cognitive coordinate system. It takes the user's current cognitive coordinates as the starting point, the group knowledge potential energy map as the navigation map, the course cognitive load index as the slope reference, and the course knowledge bridging value as the path selection basis. It conducts multi-step forward-looking path exploration, generates the lowest cognitive energy path from the current state to the target knowledge area, and outputs the course sequence mapped by the path as the final recommendation result.

[0041] It should be specifically noted that the adaptive guidance path generation module sorts the multiple candidate paths explored using a comprehensive scoring method. The scoring indicators include path cognitive energy consumption, predicted goal achievement rate, historical path satisfaction, and resource diversity. The path with the highest comprehensive score is selected as the optimal path for recommendation, and a detailed recommendation result is generated, including resource name, learning time suggestion, cognitive load prompt, and learning method guidance.

[0042] It should be further explained that, with "precise navigation" as the goal, four types of core data are selected. The data requested by users in real time includes the user's clear learning objectives (such as "mastering the application of the Pythagorean theorem"), the current learning scenario (preview / review / exam preparation), time constraints (such as "completing the learning within 1 hour"), and personalized preferences (such as "preferring video learning").

[0043] The personal cognitive coordinate system acquires real-time data on the user's current cognitive coordinate value, cognitive decay coefficient, and recent cognitive state change trends (such as the rate of increase / decrease in cognitive depth).

[0044] The group knowledge potential energy map data includes the potential energy value of each knowledge unit, knowledge association paths, and resource node distribution (cognitive load, bridging value).

[0045] Historical path feedback data extracts users' past learning effects after using the guided path (such as goal achievement rate, improvement in cognitive depth), path satisfaction evaluation (such as "path too long" or "resources unsuitable"), and path adjustment records (such as users voluntarily skipping a resource).

[0046] As attached Figure 5 As shown, a two-stage processing approach of "intent recognition + goal decomposition" is adopted. Natural Language Processing (NLP) is used to parse the learning objectives in the user's real-time request and map them into the target knowledge region in the group knowledge potential energy map. Based on the gap between the individual's cognitive coordinate system and the target region, the overall goal is decomposed into several sub-goals (e.g., the overall goal "mastering the application of the Pythagorean theorem" is decomposed into "understanding the definition of the Pythagorean theorem → mastering basic calculations → solving simple application problems → solving comprehensive application problems"). A "goal feasibility assessment" mechanism is introduced, which combines the user's time constraints and cognitive improvement rate to determine whether the goal is achievable. If it is not achievable, adjustment suggestions are given (such as extending the time or reducing the difficulty of the sub-goals).

[0047] With the goal of "minimizing cognitive energy consumption," a "multi-constraint path optimization model" is constructed, introducing cognitive energy consumption calculation. Cognitive energy consumption is defined as the sum of the energy consumption of each resource in the path, and the formula for calculating the energy consumption of a single resource is as follows: Total energy consumption along the path: ;in, For the first Cognitive energy consumption of each resource No. Cognitive load index of each resource For the first The estimated learning time for each resource. This represents the cognitive decay coefficient for the corresponding knowledge unit. The formula quantifies the energy consumption cost of different paths, representing the number of path resources.

[0048] An improved A* algorithm is adopted, starting from the current cognitive coordinates, ending at the target knowledge region, and using the group knowledge potential graph as the map to explore a path 3-5 steps forward, predicting subsequent changes in cognitive state and avoiding "short-sighted paths". Time constraints (total path duration ≤ user-set time), preference constraints (prioritizing resources of user preference type), and bridging value constraints (prioritizing resources with bridging value ≥ 4 points) are used as algorithm constraints to filter out paths that do not meet the requirements.

[0049] The multiple candidate paths identified are ranked using a comprehensive scoring method. The scoring indicators include cognitive energy consumption (weight 40%), goal achievement rate (weight 30%), historical satisfaction (weight 20%), and resource diversity (weight 10%). The path with the highest score is selected as the optimal path, and the knowledge unit sequence in the path is mapped to the corresponding course sequence, generating detailed recommendation results including "resource name, learning time suggestion, cognitive load prompt, and learning method guidance". A "path dynamic adjustment interface" is established. If the user's cognitive state changes significantly during the learning process (e.g., the accuracy rate of a resource is only 30% after learning), the system recalculates the path in real time and pushes adjustment suggestions.

[0050] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments of this disclosure. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An educational course recommendation system based on big data, characterized in that, include: Data perception and fusion module: Embedded semantic disambiguation and context reconstruction, it collects native educational data, performs cross-modal intent extraction and context anchoring, generates behavioral atoms with unified spatiotemporal stamps and semantic contexts and stores them in a distributed file system to form a native educational data space; Individual-Group Co-evolution Module: Asynchronously reads behavioral atoms from the native education data space, performs individual learning path tracking and group knowledge topology mining, deconstructs the intensity of user input and cognitive depth in different knowledge units, constructs a personal cognitive coordinate system that reflects the dynamic knowledge state, analyzes the co-occurrence and transmission patterns between group behavioral atoms, and dynamically draws a group knowledge potential energy map. Course Resource Labeling and Organization Module: This module analyzes course resources in the native educational data space, examines the mapping relationship between course content and the group knowledge potential energy map, and analyzes the effect feedback after consumption by users at different cognitive coordinate points. It labels the dynamic cognitive load index and knowledge bridging value, embedding them into the group knowledge potential energy map in the form of nodes. The adaptive guidance path generation module receives real-time requests from users and their corresponding personal cognitive coordinate system. It takes the user's current cognitive coordinates as the starting point, the group knowledge potential energy map as the navigation map, the course cognitive load index as the slope reference, and the course knowledge bridging value as the path selection basis. It conducts multi-step forward-looking path exploration, generates the lowest cognitive energy path from the current state to the target knowledge area, and outputs the course sequence mapped by the path as the final recommendation result.

2. The education course recommendation system based on big data according to claim 1, characterized in that: The native educational data includes multimodal learning behavior data, context-related data, and cognitive auxiliary data; the semantic disambiguation is based on the knowledge graph and behavioral semantic dictionary in the education field, and adopts a pre-trained model enhanced by attention mechanism, which integrates attention and emotion features in cognitive auxiliary data to perform semantic parsing of ambiguous behaviors; the context reconstructor binds different modal data to a unified spatiotemporal stamp through a spatiotemporal alignment algorithm, and adopts a dynamic allocation mechanism of context weight to realize context anchoring of behaviors.

3. The education course recommendation system based on big data according to claim 1, characterized in that: The specific construction process of the personal cognitive coordinate system includes using the knowledge unit topology as the coordinate axis, the horizontal axis as the logical sequence of knowledge units, and the vertical axis as the comprehensive value determined by the cognitive depth feature set. Through a dynamic weighted factor analysis algorithm, the intensity of an individual's investment in each knowledge unit and the cognitive depth are transformed into coordinate values.

4. The education course recommendation system based on big data according to claim 3, characterized in that: The intensity of investment is calculated by combining the proportion of learning time, the number of repeated learning sessions, and the frequency of active interaction. A cognitive decay coefficient is introduced to compensate for the decay of historical data. The cognitive depth is quantified by mapping behavior type to cognitive target classification level and combining answer accuracy and question resolution efficiency indicators.

5. The education course recommendation system based on big data according to claim 1, characterized in that: The group knowledge potential graph is drawn using a fusion algorithm of association rule mining and community discovery. It identifies high-frequency associated knowledge unit combinations and divides knowledge communities. The knowledge potential is defined as a function of the average group cognitive depth, the proportion of people who have mastered the knowledge, and the degree of knowledge association. Then, the potential value of each knowledge unit is calculated.

6. The education course recommendation system based on big data according to claim 1, characterized in that: The cognitive load index is a dynamic indicator based on the inherent characteristics of resources and user feedback. It is calculated by comprehensively considering the number of resource knowledge points, the complexity of information presentation, the average learning time of users, the error rate, and the frequency of attention distraction.

7. The education course recommendation system based on big data according to claim 1, characterized in that: The knowledge bridging value is an efficiency indicator of how well resources help users reach a target cognitive state from their current state of understanding. It is calculated by combining the matching degree between resources and the group's knowledge potential map with the efficiency of the improvement in the user's cognitive depth after using the resources.

8. The education course recommendation system based on big data according to claim 1, characterized in that: The adaptive guidance path generation module sorts the multiple candidate paths explored using a comprehensive scoring method. The scoring indicators include path cognitive energy consumption, predicted goal achievement rate, historical path satisfaction, and resource diversity. The path with the highest comprehensive score is selected as the optimal path for recommendation, and detailed recommendation results are generated, including resource name, learning time suggestion, cognitive load prompt, and learning method guidance.

Images