Abstract
This study presents a model for the organization of educational content customized for environments of individual studies. For many students, the availability of content in a general form is often inefficient. We propose a multilevel structure of concepts to provide the development of different combinations to show the same content. The work demonstrates that it is possible to customize content to encourage students using proximal learning standards. These patterns are obtained from the analysis of students with positive results in individual content organization. The formal representation establishes the student profile, multi-level content, and the distribution plan for concept correction and teaching trajectory. The trajectory structure is formally established using the method of finite differences. The system utilizes artificial intelligence techniques to organize and personalize content reactively. Customization is provided by an artificial neural network (MLP) that classifies student profiles and assigns them to a standard proximal learning pattern. To mediate and adjust the contents, a reactive system was implemented using a set of rules from teaching experts. Experimental results indicated the suitability of the approach, showing that the intelligent system efficiently established custom content structures, resulting in higher final averages and lower study times.
Introduction
The rapid expansion of Distance Learning (EAD) has brought to light a significant pedagogical challenge: the static nature of digital content. Most online platforms deliver a "one-size-fits-all" curriculum that ignores the diverse prior knowledge, learning speeds, and psychological temperaments of students. Unlike a physical classroom where a teacher can intuitively adjust their explanation based on a student's confusion, digital environments often lack this reactive mediation. This research addresses the challenge of building an Intelligent Tutoring System (ITS) capable of simulating human pedagogical intuition to provide a truly personalized learning experience.
Methodology
The proposed model is a hybrid system that combines connectionist AI (Neural Networks) with symbolic AI (Expert Rules) to create a dynamic and adaptive learning environment.
Multilevel Content Structure
Instead of a linear sequence of pages, each educational concept is authored at five distinct levels:
- Facilitated: Simplified explanations for beginners.
- Medium: Standard instructional material.
- Advanced: In-depth analysis for high-performing students.
- FAQs: Common questions and answers.
- Examples: Practical applications of the concept.
Student Profiling ($\Omega$)
The system identifies the student using a vector of 17 variables:
- Psychological Profile: 8 variables derived from the Keirsey Temperament Sorter, based on Jungian personality types.
- Contextual Factors: 7 variables regarding technological skills and socio-cultural background.
- Prior Knowledge: 2 variables from a subject-specific pre-test.
Neural Network and Proximal Learning
A Multi-Layer Perceptron (MLP) neural network takes these 17 inputs and predicts a "Proximal Learning Pattern" ($L_\Omega$). This pattern represents the probability that a specific content level will be effective for a given student. This approach is grounded in Lev Vygotsky’s "Zone of Proximal Development" (ZPD), which focuses on the gap between what a learner can do without help and what they can do with support.
Trajectory Correction via Finite Differences
To manage the student's progress, the system employs the Finite Difference Method, a numerical technique typically used in physics. This allows the system to model the student's "trajectory" through the content as a mathematical path that can be corrected in real-time if the student deviates from the optimal learning curve.
Hybrid Decision Engine
To ensure pedagogical integrity, the system combines the Neural Network's predictions with a set of Expert Rules derived from 27 experienced professors. A Monte Carlo probabilistic selection method is used to choose the next best step for the student, ensuring the system remains flexible yet grounded in sound teaching logic.
Experimental Results
The system was evaluated using a custom dataset involving 228 university students. The experiment compared "Intelligent Navigation" (the proposed model) against "Free Navigation" (student choice) and "Random Navigation" control groups.
The results demonstrated the clear superiority of the intelligent approach:
- Academic Performance: Students using the intelligent system achieved a higher average final grade (7.21) compared to those in free navigation (6.87) and random navigation (5.93).
- Efficiency: The intelligent group reached these higher scores while viewing 25% less content and spending 30% less time (26.8 minutes vs. 37.88 minutes for the free navigation group).
- Navigation Coherence: The intelligent system achieved 0% navigation incoherence, meaning it never suggested advanced content to a student who had not yet mastered the foundational concepts.
Conclusion
The research demonstrates that a hybrid AI model using proximal learning patterns can effectively automate the personalization of educational content. By integrating neural networks with expert rules and mathematical trajectory modeling, the system provides a reactive environment that significantly improves student performance and efficiency. Future work will focus on applying this model to more abstract subject areas and exploring its potential as an assistive technology for students with specific learning disabilities.
Summary updated