Dynamic science popularization project management system based on interdisciplinary data fusion

Through interdisciplinary data fusion and cognitive modeling, a dynamic science popularization project management system is built, and the problems of subject isolation and feedback lag in traditional science popularization teaching are solved, personalized teaching and efficient resource scheduling are realized, and learning effect and resource utilization efficiency are improved.

CN120450635APending Publication Date: 2025-08-08ZHEJIANG JINGHANG SHUREN CULTURE MEDIA CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202510553066.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-29
Publication Date
2025-08-08

AI Technical Summary

Technical Problem

Traditional popular science teaching systems have problems such as subject isolation, static tasks, delayed feedback and extensive resource scheduling, which is difficult to support interdisciplinary in-depth research and personalized learning.

Method used

Build a multimodal knowledge representation space, realize interdisciplinary data fusion through heterogeneous graph embedding networks and improved knowledge graphs, combine cognitive modeling and reinforcement learning strategies, and design dynamic task generation and intelligent scheduling engines to realize real-time task scheduling, personalized feedback and cognitive evaluation.

Benefits of technology

It realizes machine-perceptible task generation for interdisciplinary problems, improves the accuracy of learning efficiency, resource utilization efficiency and cognitive feedback, supports personalized teaching processes, and breaks the structural bottleneck of traditional teaching.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120450635A_ABST
    Figure CN120450635A_ABST
Patent Text Reader

Abstract

The invention discloses a dynamic science popularization project management system based on interdisciplinary data fusion. According to the system, a multi-source data warehouse is constructed, astronomy, geography, programming behaviors and evaluation data semantics are aligned into a knowledge graph, and a personalized project type exploration task is generated through semantic path search in combination with a student cognitive state graph; the multi-agent reinforcement learning scheduling engine allocates resources in real time, optimizes task paths, fuses behavior modeling and a fuzzy evaluation model in task execution to generate cognitive feedback and advanced capability evaluation, and realizes automatic task generation, intelligent scheduling and closed-loop evaluation in popular science teaching.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary technical field of educational informatization, artificial intelligence and multi-source heterogeneous data fusion, and specifically relates to a data-driven teaching management system that supports interdisciplinary project-based learning, especially a dynamic science popularization project management system that integrates astronomical observation data, geographic space information, programming behavior trajectories and subject evaluation feedback. It is widely used in intelligent teaching scenarios such as primary and secondary school science education, science popularization research and study institutions, and science popularization exhibition halls. It has educational adaptability, task intelligence and cognitive drive characteristics, and belongs to the key technology category of the new generation of intelligent education infrastructure and cognitive enhancement systems. Background Art

[0002] With the gradual transformation of the education system under the "double reduction" policy and the guidance of core competencies, project-based learning (PBL) and interdisciplinary integrated teaching have gradually become important models of popular science education in primary and secondary schools. Traditional popular science teaching relies heavily on linear curriculum design and static teaching materials. It lacks the ability to multi-dimensionally model complex real-world problems and cannot effectively support students in conducting in-depth exploration in the context of real data and interdisciplinary backgrounds. In addition, most current popular science platforms still have obvious limitations in terms of technical implementation: on the one hand, there is widespread structural heterogeneity, semantic fragmentation, and update delays among various types of educational data, which makes it impossible for the system to build a unified cognitive model and dynamic task logic; on the other hand, individual students differ in cognitive level, interest orientation, and task execution ability. Traditional systems lack effective dynamic scheduling mechanisms and the ability to build personalized learning paths, often presenting a management dilemma of "fixed tasks, delayed feedback, and rough evaluation."

[0003] Although existing project management platforms have certain process control functions, their core is still limited to static resource release and manual evaluation drive. They are unable to realize dynamic task generation based on real-time data fusion, and lack the ability to fine-grainedly model student behavior trajectories and learning preferences. Some systems that attempt to introduce artificial intelligence technology often only stay at the surface recommendation level, failing to achieve full-link intelligent integration from data collection, knowledge reconstruction to task-driven, resulting in insufficient intelligence in teaching behavior and inefficient utilization of educational resources, making it difficult to support true interdisciplinary exploration and cognitive transfer.

[0004] In this context, there is an urgent need for an intelligent system that integrates cutting-edge data intelligence technology and cognitive modeling strategies, and can realize real-time task scheduling, dynamic learning path generation and personalized feedback mechanism based on multimodal data, so as to provide technical support for the entire process of scientific education "data-driven-intelligent recommendation-project generation-dynamic evaluation-cognitive advancement", and fundamentally solve the key technical problems of traditional popular science teaching in structural organization, content generation and intelligent adaptation. Summary of the Invention

[0005] The present invention innovatively proposes an educational data fusion system architecture driven by interdisciplinary semantic modeling, breaking through the fundamental limitations of traditional teaching systems in terms of closed disciplines, data fragmentation, and static tasks. By constructing a unified multimodal knowledge representation space, the system restructures and semantically aligns multi-source heterogeneous educational data such as astronomical observation images, geographic information system (GIS) data, graphical programming behavior trajectories, experimental interaction process data, and cognitive assessment results, forming a general interdisciplinary data middle platform with strong expression capabilities and computability. In the underlying data fusion mechanism, the present invention introduces a new type of heterogeneous graph embedding network, combined with improved knowledge graph construction technology, to perform entity abstraction, relationship deconstruction, and semantic co-embedding of different data modalities, realizing deep fusion and linkage analysis of interdisciplinary data in a unified graph space.

[0006] Compared to existing systems that handle educational data in isolation, this invention achieves a first-of-its-kind, closed-loop system architecture, encompassing data unification, knowledge collaboration, and task generation. Its unique feature is its use of the data warehouse itself as not only a storage center but also a semantic engine node for dynamic task generation and cognitive mapping. The system's data model is self-evolving, updating data weights and knowledge graph connections in real time based on student interactions. This enables the system to automatically reason and construct interdisciplinary task modules for diverse disciplinary tasks based on a multi-dimensional matrix of semantic proximity, historical behavior, and ability prediction. This architectural design enables machine-perceivable and task-computable cross-disciplinary problem spaces in science popularization education for the first time. This not only enables the intelligent reorganization and adaptive delivery of educational resources, but also provides the theoretical support and technical foundation for building next-generation intelligent educational systems capable of cognitive evolution. This proposed system architecture pioneers a new intelligent education technology approach based on the three-dimensional framework of "knowledge graph + behavioral data + task logic," demonstrating significant originality and cutting-edge breakthrough value.

[0007] Based on the unified modeling and semantic alignment of multimodal data, this paper further proposes a dynamic task generation mechanism that integrates cognitive modeling and reinforcement learning strategies, and designs a supporting intelligent scheduling engine with high-dimensional multi-objective optimization. This enables the system to achieve optimal coordination between personalized task construction and resource allocation in project-based teaching scenarios. This mechanism, for the first time, uses the "student cognitive state map" as the core variable of task driving, thereby breaking through the traditional teaching system's single arrangement logic based on static course packages and unified tasks as the path, and instead forms a closed-loop generation paradigm based on "cognitive differences-task decomposition-path fitting-intelligent scheduling". Specifically, the task generation module integrates a semantic reasoning network based on the graph attention mechanism (GAT) and a Transformer architecture for sequence modeling. It performs path search and semantic completion in the knowledge graph, performs high-dimensional matching between the current student knowledge state and task requirements in the graph, and automatically generates task themes, related subject content nodes, required resource elements, and time-ability mapping functions. This ensures that the task design is cross-disciplinary in content, dynamically controllable in difficulty, and goal-oriented in structure. The system introduces a "learning load prediction module" based on cognitive load theory, which is used to judge students' learning fatigue curve and cognitive saturation point in continuous task execution. Through the bidirectional GRU network, the behavior sequence is predicted and analyzed, making the task push process more in line with the students' rhythm and rhythm, improving learning efficiency and maintaining exploration motivation.

[0008] The accompanying intelligent scheduling engine, based on a multi-objective evolutionary optimization framework and integrating multi-agent collaborative scheduling strategies from reinforcement learning (MARL), designs a scheduling reward function system that combines resources, paths, and feedback. After task generation, this system dynamically plans task delivery paths based on factors such as the student's current available time, device accessibility, task dependencies, and historical learning outcomes. The intelligent resource matching module integrated within the scheduling engine adapts various resources, including virtual lab platforms, image simulation data, physical kits, and programming environments, to teaching objectives, achieving the three goals of precise resource distribution, optimized spatiotemporal allocation, and dynamic load balancing. Furthermore, the system's scheduling layer incorporates a fuzzy logic fusion mechanism to enable robust control in scenarios with imprecise feedback. For example, in scenarios where student feedback data is missing or of insufficient quality, the system can estimate task migration based on clustered paths from similar cognitive maps, maintaining project progress continuity and appropriate task depth, enhancing the system's fault tolerance and generalization capabilities in real-world teaching environments. Experimental data shows that this scheduling mechanism can improve resource utilization efficiency by 28.7% and student task completion rate by 34.2% while maintaining a dynamic balance of average task difficulty, which is better than the traditional static scheduling mode.

[0009] This task generation and intelligent scheduling mechanism not only expands the functionality of existing teaching systems but also revolutionizes the depth of intelligent teaching and the breadth of cognitive matching. Through the technical approach of "cognitive state modeling + semantic graph reasoning + multi-agent reinforcement learning," it achieves a paradigm shift from "unified construction" to "individual drive" for interdisciplinary tasks, creating a new model for adaptive intelligent regulation of educational systems, demonstrating both strong technological foresight and unique application.

[0010] To build a complete intelligent closed-loop teaching system, this paper introduces a behavioral analysis and feedback evaluation mechanism for cognitive advancement after task execution. This innovatively integrates key technologies such as experimental behavior modeling, feedback semantic deconstruction, and multimodal data attribution to form an intelligent evaluation system that integrates behavioral process evaluation, task performance evaluation, and cognitive outcome evaluation. This mechanism not only addresses the limitation of traditional education systems that "only evaluate results, not processes," but also enables the system to automatically extract student cognitive transition information under multi-task, multi-stage, and multi-path conditions, providing a refined decision-making basis for task generation and path adaptation in the next stage.

[0011] In terms of experimental behavior modeling, this paper proposes a Multi-Granular Behavioral Semantic Network (MGBSN). This network encodes low-level behavioral data such as clickstreams, programming module combinations, image interaction gestures, and device operation paths during student task completion into high-dimensional semantic behavior sequences. Using dynamic time warping and behavioral feature clustering techniques, it identifies key behavioral nodes with cognitive signatures, such as "concept transfer attempts," "experimental branch construction," and "error fallback strategies." Based on this, the system constructs a cognitively explanatory individual behavior trajectory graph (Cognitive Behavioral Graph), providing a data foundation for the evaluation model. Regarding the feedback mechanism, the system incorporates a dual-channel semantic deconstruction module: First, it performs structural semantic analysis on student-submitted task outputs (text reports, graphic artwork, code snippets, etc.), extracting the relevance of core knowledge points, logical rigor, and innovation index; second, it analyzes students' emotional vocabulary, metacognitive representations, and strategy expressions in feedback questionnaires or self-descriptions, thereby constructing a dual feedback vector of "results + subjectivity." The feedback information is updated through the dynamic knowledge state estimation network (Dynamic Knowledge State Estimator, DKSE) designed by the present invention, and drives the iterative optimization of the individual knowledge graph together with its historical learning trajectory. In terms of evaluation decision-making, the system adopts a multi-factor fuzzy hierarchical evaluation model (Fuzzy Multi-Factor Evaluation Model, FMFEM), which comprehensively considers multiple indicators such as task completion quality, behavioral path rationality, resource call efficiency, cognitive span value, and constructs a set of dynamic evaluation vectors for multi-dimensional evaluation and visualization presentation. More importantly, the present invention designs an evolution mechanism for the cognitive ability index (Cognitive Ability Index, CAI), which can depict the cognitive growth trend of students in real time according to variables such as the behavioral complexity, knowledge transfer path length and strategy innovation frequency in projects at different stages, and provide teachers with an operational data view with academic explanatory power and intervention guidance significance. In real-world deployments, the system demonstrated excellent behavioral recognition accuracy (92.3%) and cognitive stage consistency (Kappa coefficient of 0.81). In teacher intervention scenarios, it assisted teachers in identifying implicit learning barriers such as strategic stagnation and insufficient conceptual leaps, effectively improving teaching adaptability and individual response speed. Compared to traditional single-point evaluation methods based on test scores, this system achieves a continuous cognitive perception and feedback loop, representing a paradigm shift in traditional educational evaluation mechanisms.

[0012] In summary, the present invention integrates cognitive behavioral modeling, semantic feedback analysis and intelligent evaluation logic into the dynamic education process, which not only realizes the deep identification and continuous tracking of students' learning status, but also promotes the fundamental transformation of the education system from "external observation" to "internal understanding", improves the level of teaching personalization, feedback interpretation ability and cognitive intervention accuracy, and is a groundbreaking technical path innovation in the current field of intelligent education. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0015] Attachment Figure 1 This is the overall architecture diagram of the dynamic science popularization project management system based on interdisciplinary data fusion.

[0016] Attachment Figure 2 It is a flowchart for the fusion of multi-source heterogeneous educational data and the construction of semantic graphs.

[0017] Attachment Figure 3 It is a task generation and scheduling flowchart based on cognitive modeling and intelligent optimization.

[0018] Attachment Figure 4 It is a path diagram for student behavior modeling and cognitive feedback evaluation.

[0019] Attachment Figure 5 This is a diagram of system performance comparison experiment results. DETAILED DESCRIPTION

[0020] 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 them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationships and movement of various components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indicators will also change accordingly.

[0022] In addition, the descriptions of "first," "second," etc. in the present invention are for descriptive purposes only and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features specified as "first," "second" may explicitly or implicitly include at least one of such features. The technical solutions between the various embodiments can be combined with each other, but they must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0023] As the current education system comprehensively promotes interdisciplinary integration, inquiry-based learning, and intelligent education platforms, traditional science popularization project teaching management systems are plagued by issues such as disciplinary isolation, static tasks, delayed feedback, and extensive resource allocation, making them unable to meet the personalized development needs of the new generation of students in terms of cognitive development and inquiry paths. This is particularly true in application scenarios such as science popularization teaching in primary and secondary schools, research institutions, and science museums, where learning tasks often require dynamic adaptation to the knowledge background, learning abilities, and behavioral characteristics of different fields. This places extremely high demands on the system's real-time responsiveness and intelligent task-driving capabilities.

[0024] To this end, this paper proposes a dynamic science popularization project management system based on interdisciplinary data fusion. Through a unified multimodal data warehouse, an interdisciplinary knowledge graph construction mechanism, a cognitive state-driven task generation model, and a reinforcement learning scheduling engine, it achieves closed-loop intelligent management and control of the entire process, from task design and execution, feedback, to evaluation. The system automatically integrates multi-source heterogeneous educational data, including astronomical observation images, geospatial information, programming behavior trajectories, experimental interaction processes, and evaluation results. It constructs a unified cognitive space through adaptive semantic modeling and graph embedding technology, and on this basis generates dynamic exploration task paths, achieving precise task matching, optimal resource scheduling, and a personalized teaching process driven by cognitive feedback.

[0025] Next, we will describe various embodiments of the present invention in detail. These embodiments, combined with the accompanying drawings, will showcase key technical details from cross-modal data acquisition, knowledge graph construction, intelligent task generation and scheduling, to behavioral process modeling and cognitive evaluation feedback, fully demonstrating the advantages and broad application prospects of the present invention in terms of structural design, system linkage, technological innovation, and practical application.

[0026] Example 1

[0027] This embodiment provides an overall architecture design of a dynamic science popularization project management system based on interdisciplinary data fusion. Figure 1As shown in the figure, a four-layer semantic-driven system structure is adopted, namely the user layer (S101), the task-driven layer (S102), the data fusion layer (S103) and the scheduling feedback layer (S104). The four layers operate in coordination through a unified data channel and semantic logic interface, forming a generalizable, evolvable and cognizable dynamic science popularization teaching platform.

[0028] The outermost layer of the system is the user interaction layer (S101). Its core lies in building a multi-role, multi-channel task portal, supporting student terminals (S10101) to instantly access task modules, experimental platforms, and feedback channels, and providing the teacher console (S10102) with a complete project planning, path supervision, and personalized task publishing interface. The system further expands scenario adaptability through the research and study institution access module (S10103), covering off-site teaching environments such as science museums and education bases. Below this user layer, the task-driven layer (S102) constitutes the intelligent generation core of the system, which integrates the cognitive state perception module (S10201) to perceive the students' ability state map and task history behavior chain in real time; the task generation engine (S10202) automatically assembles individualized exploration tasks through graph neural networks and semantic path search mechanisms, and the embedded task decomposition module (S10202a) supports the decomposition of complex interdisciplinary problems into computable knowledge path units, and ensures that the task content maintains a dynamic balance in knowledge span, subject distribution and resource density through the subject fusion controller (S10202b).

[0029] After task generation is complete, the task semantic scheduling module (S10203) plays a key role. The system uses it to jointly model task execution priorities, resource status, and student cadence, and pushes the scheduling results to the underlying data fusion layer (S103). This fusion layer continuously aggregates multi-source data from astronomical observation platforms, geographic information systems, programming environments, and assessment engines through the data acquisition module (S10301). The cross-modal feature fusion network (S10302) completes semantic alignment and feature mapping, and establishes a dynamically updated knowledge graph model in the graph construction engine (S10303). The entity abstraction unit (S10303a) is responsible for high-level concept modeling, while the relationship connector (S10303b) is used to build upstream and downstream knowledge links between different subject entities.

[0030] The bottom-level intelligent scheduling and feedback layer (S104) implements the closed-loop intelligent decision-making function of the entire system. Its built-in multi-agent scheduling engine (S10401) continuously optimizes the adaptation efficiency of tasks and resources through reinforcement learning mechanisms, and perceives the platform's operating load through the resource status monitoring module (S10402), ensuring system stability and task delivery quality in large-scale concurrent scenarios. The cognitive feedback evaluation module (S10403), as the terminal node of the system's cognitive loop, combines the behavior analysis network (S10403a), the capability evolution estimator (S10403b), and the teacher intervention suggestion generator (S10403c) to construct a multi-layered cognitive evaluation system that combines subjective feedback and objective behavior path analysis.

[0031] To sum up, this embodiment, through a system architecture that combines hierarchical module division with a semantic coupling mechanism, has for the first time realized a dynamic project task generation mechanism driven by individual students' cognitive states, and opened up the full process path of "data-knowledge-task-feedback" at the structural level. It not only greatly improves the composability and reuse efficiency of interdisciplinary teaching resources, but also provides a general technical framework with cognitive evolution capabilities for the construction of intelligent education systems. Its systematic original design is the first of its kind in the existing teaching platform architecture.

[0032] Example 2

[0033] This embodiment provides a path to implement the fusion of multi-source heterogeneous educational data and the construction of interdisciplinary semantic graphs in the system. Figure 2 As shown, the entire technical process runs through the complete link from raw data collection, structured representation generation, to knowledge semantic embedding and graph structure construction, aiming to break through the semantic barriers between different modalities, different disciplines, and different cognitive levels, and build a unified knowledge expression framework with real-time update capabilities and structural self-evolution characteristics to support deep semantic support for subsequent task generation, scheduling, and cognitive evaluation.

[0034] This embodiment first gathers five core educational data sources at the input end, namely astronomical observation image data (S20101), geographic space vector data (S20102), graphical programming behavior log (S20103), subject assessment and feedback results (S20104) and experimental process trajectory data (S20105). These data show structural heterogeneity, temporal asynchrony and semantic level unevenness, which are difficult to be directly processed by traditional data platforms. In this system, all raw data first enter the preprocessing module (S202), which includes a structural standardization unit (S20201) and a feature aligner (S20202). The former uniformly reshapes the dimensions and normalizes the formats of images, vectors and text data, and the latter uses a time synchronization mechanism and a context normalization strategy to convert different modal data into embedded expressions with a consistent timeline and a unified unit space.

[0035] After data normalization, all data enters a unified embedding process via the semantic fusion module (S203), whose core components include a heterogeneous graph embedding encoder (S20301), an inter-modal semantic alignment optimizer (S20302), and a shared representation space generator (S20303). The system employs a combined modeling strategy of graph neural networks and attention mechanisms to ensure not only a linear mapping between modalities in terms of feature dimensions but also a deep alignment of semantic structures. This allows the "magnitude variation" in astronomical images to be logically linked to the "photoperiod" in geographic systems, forming a cross-disciplinary nested linkage with the "physical modeling variables" in programming tasks.

[0036] After semantic alignment, the system inputs all embedding vectors into the graph construction module (S204). The entity node abstractor (S20401) constructs unified educational object knowledge entities (such as "celestial body," "geomorphology unit," "motion formula," and "student cognitive concept unit"). The relationship chain builder (S20402) automatically identifies and defines multidimensional and polymorphic relationships between entities, including edge structures such as "inclusion," "triggering," "dependency," and "equivalent migration." Finally, the knowledge graph update engine (S20403) completes multiple rounds of iterative generation and version management of the graph structure. This engine possesses continuous learning capabilities and can adjust the node weights, edge density, and subgraph activity in the graph in real time based on student task execution data and behavioral path feedback. This allows the system's knowledge structure to evolve dynamically during the teaching process, supporting a semantically driven teaching approach where "structure evolves with cognition."

[0037] The data fusion path proposed in this embodiment breaks through the cognitive model bottleneck of "content isolation and fixed structure" in traditional teaching platforms, and realizes the highly consistent fusion expression of multimodal, multi-level, and cross-disciplinary educational data in the same semantic graph. It realizes the structural unification of "subject entity-learning behavior-cognitive mapping" at the system level for the first time, and is a basic technical route innovation to promote the future intelligent education platform to achieve deep semantic support.

[0038] Example 3

[0039] This embodiment provides a dynamic task generation and intelligent scheduling method based on cognitive modeling and reinforcement learning optimization strategy in the system of the present invention, such as Figure 3 As shown in the figure, the process integrates four major technical paths: semantic reasoning, temporal modeling, capability perception, and resource scheduling. It aims to achieve highly personalized construction of task content and intelligent global optimization of execution paths, and ultimately achieve the ideal state of "tasks vary according to people, and resources are matched on demand" in project-based teaching.

[0040] The system begins with the cognitive state input module (S301). Taking each student as a unit, it constructs a current cognitive state map based on multi-dimensional features such as historical learning trajectories (S30101), real-time ability assessment values (S30102), and interest preference vectors (S30103). This map uses a graph neural network to model the topological structure of students' knowledge mastery. Combining their operational paths and strategy distribution, it predicts the likelihood of cognitive transitions in future tasks, providing precise path constraints and difficulty control for task generation.

[0041] The task generation module (S302) runs on this basis. Its semantic path search engine (S30201) performs reverse retrieval of targets based on graph entity nodes and relationship edges, and traces back potential knowledge transfer paths from cognitive targets. Subsequently, the task assembly strategist (S30202) reconstructs the path nodes into an inquiry task structure, and generates a stage task chain by combining the teaching objectives and the task granularity model. The task executable mapper (S30203) then converts the semantic path into an interactive task object for the student terminal. This module has a knowledge point coverage controller (S30203a) and a stage target detector (S30203b) to ensure that the task is both cognitively progressive and accessible.

[0042] After task paths are generated, the scheduling strategy module (S303) combines the state of teaching resources with students' cognitive cadence to implement multi-objective optimization scheduling. The scheduling unit (S30301) within this module employs a combined modeling approach of evolutionary learning and reinforcement learning to construct a three-dimensional optimization objective: time, resources, and feedback. The multi-agent learning controller (S30302) dynamically coordinates the execution paths of multiple students' concurrent tasks to avoid resource conflicts and cognitive saturation. The cognitive load monitoring module (S30303) tracks students' task execution cadence and mental load in real time, automatically adjusting task push frequency and resource granularity.

[0043] The system sends the generated task path to the student terminal and platform interface through the result output and feedback module (S304), where the task distributor (S30401) binds individual accounts and specifies the resource mapping generator (S30402) to match them with various supporting resources such as required experiments, image simulations, and data analysis. The scheduling result log generator (S30403) automatically generates a multi-dimensional task allocation report for teacher supervision, evaluation, and secondary intervention.

[0044] The task generation mechanism described in this embodiment is not a simple recommendation or content splicing, but a semantically driven task construction logic built on the basis of cognitive modeling and graph reasoning, and completes intelligent scheduling under the multi-agent optimization framework. For the first time, it realizes the full-process data closed loop from knowledge graph-cognitive state-task generation-scheduling execution in the education platform, and constructs a new intelligent teaching path with "cognitive perception-task reconstruction-resource adaptation". Its technical model is still blank in similar systems, and it has extremely high creativity and practical application value.

[0045] Example 4

[0046] This embodiment provides a dynamic evaluation mechanism based on the integration of behavior modeling and cognitive feedback, which realizes the continuous perception and intelligent interpretation of students' learning paths, operation behaviors, feedback content and cognitive status. Figure 4 As shown in the figure, the path diagram constructs a complete closed-loop cognitive analysis process through four stages: behavioral data collection (S401), process modeling (S402), semantic feedback (S403) and cognitive evaluation (S404). It breaks the coarse-grained management model of traditional teaching systems that can only "evaluate after the fact" and "examine the results", and realizes the real-time semantic analysis of the learning process and the structural quantification of the individual ability growth trajectory for the first time.

[0047] In the first phase, the system actively acquires multidimensional interaction data from students during task execution through the Behavior Data Acquisition Module (S401), including records of click / slide actions (S40101), graphical programming module call trajectories (S40102), and path selection and state transition data during the experimental task (S40103). All behavioral data is automatically synchronized to the Behavior Modeling Module (S402). The integrated multi-granularity behavior modeling network (S40201) encodes behavior sequences using graph convolution and attention mechanisms, generating high-dimensional behavior vectors with temporal structure and semantic density. The Behavior Semantic Graph Generator (S40202) embeds these vectors into a unified "task-operation-cognition" graph space. The Behavior-Cognition Annotation Mapper (S40203) identifies key behavioral nodes with pedagogical significance, such as "falling back after a failed attempt," "constructing a task jump path," and "independent variable modification," thereby establishing an individual behavior-cognition graph.

[0048] During the feedback phase, the system goes beyond scoring results and instead implements a dual-channel semantic parsing structure to enable the parallel interpretation of procedural and subjective feedback. The Task Outcome Semantic Analyzer (S40301) analyzes student-submitted text reports, experimental constructions, or programming results through a three-layer semantic analysis: syntax, structure, and logic. This analysis then matches the student's submitted text reports, experimental constructions, or programming results with the task knowledge path, assessing their cognitive depth and structural rationality. The Subjective Feedback Semantic Parser (S40302) processes student self-report data, questionnaire responses, task summaries, and other textual content, extracting emotions, metacognitive vocabulary, and learning strategy expressions to form a subjective mapping of cognitive states.

[0049] All feedback data is input into the dynamic knowledge state estimation network (S40303). This network integrates the aforementioned behavioral graph and feedback semantic vectors, and through temporal reasoning and state evolution modeling, generates high-level cognitive indicators such as current knowledge mastery, transferability, and strategy usage breadth. These indicators are then uniformly mapped to the cognitive ability evaluation layer (S404). In this layer, the system uses a cognitive ability index calculator (S40401) to construct a student's ability progression curve to identify their transition intervals across multiple task stages. A multi-factor fuzzy evaluation model (S40402) quantitatively evaluates task quality, behavioral rationality, resource utilization efficiency, and strategy innovation. Ultimately, the teacher intervention suggestion generator (S40403) outputs personalized intervention suggestions and periodic summary reports, providing teachers with an executable, quantifiable, and traceable cognitive auxiliary reference.

[0050] The behavior-feedback-evaluation closed loop constructed in this embodiment not only provides a deep understanding of students' current status but also dynamically projects their cognitive development path, enabling a structural leap in the teaching process from "external behavior supervision" to "internal understanding-driven." This mechanism has achieved numerous original breakthroughs in cognitive hierarchical modeling, semantic feedback parsing, and behavioral graph generation, possessing significant potential for widespread adoption. It marks a key step in the intelligent education platform's transition from surface data collection to deeper cognitive understanding.

[0051] Example 5

[0052] This example demonstrates the performance of the system in a real teaching environment. Figure 5The comparative experimental results shown fully verify the system's system-level advantages in multiple indicators such as cognitive improvement, task response, resource scheduling, and feedback quality. The experiment adopted a control group design, using three types of educational platforms as performance baselines: traditional static teaching systems (i.e., teacher-led, task preset, offline experiments), current mainstream project-based learning platforms (i.e., with task release and subject integration functions but lacking a dynamic scheduling mechanism), and the interdisciplinary intelligent project management system proposed in this invention. Under the conditions of unified teaching objectives and student size, the degree of cognitive leap, task completion quality, and system operation efficiency induced by each platform in three consecutive rounds of exploratory tasks were compared.

[0053] The experiment involved 420 students from five middle schools, randomly divided into control and experimental groups. Each group underwent two rounds of daily, interdisciplinary tasks over a two-week period. Tasks ranged from predicting meteor shower intensity based on open-source astronomical data, to writing weather simulation code to predict storm paths, to analyzing species migration trends by combining geographic distribution with biological data. These tasks required the mobilization of multiple modal data and high-level knowledge graphs.

[0054] In terms of cognitive growth rates, this system, through cognitive state graph modeling and knowledge transfer path control, enables students to develop a more targeted and structured knowledge accumulation path within each round of tasks. Based on standardized ability assessments and pre- and post-test differences, the experimental group's average cognitive growth reached 32.6%, far exceeding both the project-based platform (18.9%) and the traditional system (12.4%). Furthermore, the system demonstrated cross-level transitions from "understanding to synthesis" and "from application to creation" in Bloom's cognitive taxonomy, demonstrating its ability to reshape cognitive structures.

[0055] In terms of task completion efficiency, this invention utilizes a multi-objective scheduling system driven by reinforcement learning to effectively regulate task execution cadence, pre- and post-process dependencies, and resource response rates. Statistics show that task timeout rates have been reduced from 34.7% in traditional systems to just 6.8% in this system, while average task response latency has been reduced from 19.2 seconds to 5.7 seconds. Concurrent processing capacity in high-frequency task scenarios has been increased by over 60%. Across all assigned tasks, the completion rate reached 84.2%, a nearly 35 percentage point improvement over traditional systems.

[0056] Regarding resource utilization efficiency, the system dynamically allocates available tools (experimental platforms, simulation software, image simulators, and data packages) based on graph semantic tags and resource demand mapping algorithms, automatically avoiding low-frequency resource redundancy. The real-time resource scheduling ratio reached 91.6%, with a redundancy rate below 7.3%. In contrast, traditional systems exhibit a severe imbalance in resource call frequency, with the gap between the highest and lowest resource usage exceeding 4.2 times, and the project platform redundancy ratio still as high as 18.6%.

[0057] Additional collected data also shows that the system has the ability to optimize the frequency of teacher intervention. In a unit task cycle, the traditional system requires each teacher to intervene in task status 9.1 times on average, while the system reduces this to 3.7 times, with a burden reduction ratio of 59.3%. The overall intervention suggestion matching rate of the platform is 92.4%, that is, the proportion of intelligent intervention suggestions generated by the system that are adopted and implemented by teachers, showing that the feedback suggestion system of the present invention (see Figure 4 Embodiment) has high decision-making credibility and teaching operability.

[0058] The above performance indicators demonstrate that the proposed system has established a dynamic educational infrastructure at a higher level than existing technologies in terms of cognitive modeling accuracy, task pacing control, resource scheduling intelligence, and feedback decision reliability. This capability stems from the high-fidelity semantic mapping of heterogeneous information by its data fusion mechanism, the interpretable constructibility of task generation paths, and the scheduling model's globally optimal control strategy for system states. Figure 5 The performance data shown is not the result of local functional optimization, but rather the result of systematic innovation in the integration of structural logic, technical path and cognitive theory of the present invention.

[0059] (1) Cognitive State Graph: This refers to a graph structured around a student's current knowledge points, ability nodes, cognitive levels, and concept migration paths. It dynamically represents their knowledge mastery and cognitive evolution. In this paper, this graph is generated by integrating historical task behaviors, real-time interaction data, and feedback evaluation results. It serves as input to the task generation engine to support personalized task construction.

[0060] (2) Task Semantic Trajectory: This refers to a task organization path that is derived by the system based on the cognitive state map and knowledge map, and has logical coherence and interdisciplinary integration characteristics. This path is composed of several semantic nodes (such as exploration goals, subtasks, key knowledge points) and directional edges (indicating sequential dependencies or knowledge transfer). It can adapt to students' cognitive rhythm and guide them to complete a task sequence with knowledge transition significance.

[0061] (3) Multimodal Educational Data: This refers to a collection of data with heterogeneous sources and diverse structures, including images (such as astronomical observation maps), spatiotemporal data (such as geographic data), behavioral data (such as clickstreams and programming behaviors), and text data (such as assessment feedback and task descriptions). The system achieves unified modeling of this type of data through feature embedding and semantic alignment mechanisms, breaking through the structural bottlenecks of traditional systems at the level of disciplinary boundaries and data silos.

[0062] (4) Knowledge Graph Construction and Evolution: This is one of the core supporting technologies of the present invention. It refers to the construction of an interdisciplinary semantic graph using entity recognition, relationship extraction, and embedding optimization technology. During the learning process, the edge weights, node status, and activity of the graph structure are adjusted in real time based on student behavior and feedback, thereby achieving continuous evolution of the graph and dynamic semantic enhancement.

[0063] (5) Reinforcement Learning-Based Scheduling Engine: This refers to the core algorithm component in the system scheduling module. By constructing a task reward function and a multi-objective optimization model of resources, tasks, and capabilities, it achieves dynamic control over the timing of task distribution, the order of resource use, and the distribution of cognitive load. The scheduling results not only consider the system's operating efficiency but also combine the individual student's task acceptance ability and cognitive load to improve the adaptability of the learning path.

[0064] (6) Cognitive Ability Index (CAI): This is a key indicator system designed for this system to dynamically quantify students’ ability development. Its calculation integrates multiple dimensions such as task performance, behavioral path, feedback content, knowledge transfer length, and strategy diversity to characterize the individual’s cognitive advancement status, providing teachers with a reliable, visual, and traceable ability assessment reference.

[0065] (7) Feedback Acceptance Rate: refers to the proportion of intervention suggestions, task adjustment prompts or stage summaries automatically generated by the system that are adopted in the teacher's decision-making process. It reflects the actual teaching applicability and trustworthiness of the intelligent intervention module and is an important parameter for evaluating the quality of system suggestions and the degree of teacher acceptance.

Claims

1. A dynamic science popularization project management system based on interdisciplinary data fusion, characterized by: The system comprises: (1) User interaction module, used to receive task requests, interaction data and feedback information from student terminals and teacher terminals; (2) Multimodal data fusion module, which is used to collect and process multi-source heterogeneous educational data including astronomical observation data, geospatial data, programming behavior data, experimental log data, and subject assessment data, and construct a unified feature representation space; (3) Knowledge graph construction module, which is used to perform semantic abstraction, entity recognition and relationship modeling on the fused multimodal data to generate a knowledge graph that supports interdisciplinary semantic reasoning; (4) Cognitive state modeling module, which is used to construct individual cognitive state maps based on students' historical behavior trajectories, learning ability indicators, and interest vectors; (5) a task generation module, which is used to perform semantic path search, task decomposition and structural reconstruction based on the cognitive state map and the knowledge map to generate interdisciplinary personalized project-based exploration tasks; (6) Reinforcement learning scheduling module, which is used to perform dynamic scheduling of task resources and multi-objective path optimization based on task dependency structure, resource status and cognitive load prediction results; (7) Cognitive feedback and evaluation module, which is used to analyze students’ operational behaviors, task results and subjective feedback during task execution, generate ability evolution assessment and intervention suggestions, and realize a closed-loop teaching process.

2. The system according to claim 1, wherein: The multimodal data fusion module includes a heterogeneous graph embedding network and a modality alignment mechanism, which supports semantic alignment and embedded expression of unstructured data such as images, texts, vectors and behavior logs.

3. The system according to claim 1, wherein: The knowledge graph construction module supports dynamic updating of the graph structure, and makes real-time evolutionary adjustments to node weights, edge connection strengths, and subgraph activity based on student task execution data and feedback results.

4. The system according to claim 1, wherein: The cognitive state modeling module constructs the topological structure of students' knowledge mastery based on graph neural networks, and integrates temporal behavior data to construct a multi-layer cognitive state representation.

5. The system according to claim 1, wherein: The task generation module adopts semantic path search and stage target detection algorithms to ensure that the generated tasks are cognitively advanced, interdisciplinary and structurally reasonable.

6. The system according to claim 1, wherein: The reinforcement learning scheduling module is based on a multi-agent reinforcement learning strategy, jointly considering task completion time, resource utilization efficiency and cognitive load distribution, and dynamically optimizing the task execution path.

7. The system according to claim 1, wherein: The cognitive feedback and evaluation module includes a multi-granularity behavior modeling network and a multi-factor fuzzy evaluation model, which are used to construct a student cognitive ability index and generate a visual cognitive advancement curve.

8. The system according to claim 1, wherein: The tasks generated by the system include task objectives, resource dependencies, time rhythm and knowledge transfer paths, and can be automatically configured through resource label matching and semantic reasoning modules.

9. The system according to claim 1, wherein: The system supports the generation and push of teacher intervention suggestions, with a matching rate of over 90%, and can be used for teacher evaluation decisions and task rhythm fine-tuning.

10. The system according to any one of claims 1 to 9, characterized in that The system can be deployed in primary and secondary school teaching scenarios, research and study activity platforms or science exhibition halls to support intelligent management and cognitive level assessment of task-driven learning.

Citation Information

Cited By

  • Method and system for evaluating digital literacy of teacher based on process data

    CN121329248A

  • Intelligent display equipment self-adaptive adjustment method for digital display

    CN122308620A