Personalizing Students' Learning Needs by a Teaching Decision Optimization Method

Yi Zhang, Yueyun Shi, Fuli BiInternational Journal of Emerging Technologies in Learning (iJET) 2023 · DOI

TL;DR

This paper introduces a hybrid AI framework that moves beyond "one-size-fits-all" education by combining Improved Ant Colony Optimization-Support Vector Regression (IACO-SVR) to predict student needs with an Improved Particle Swarm Optimization (IPSO) algorithm to optimize classroom decisions. By dynamically adjusting teaching resources and methods based on student data, the system significantly improved student understanding levels and outperformed standard optimization models.

What Problem Does This Paper Address?

Traditional classroom teaching often relies on a rigid syllabus and the personal experience of the teacher. This approach tends to treat students as a monolithic group, ignoring individual differences in knowledge acquisition, learning interests, and progress. Consequently, some students struggle to keep up while others become bored, preventing many from reaching their full potential.

The authors tackle the challenge of Personalized Education at scale. They seek to answer: How can we quantitatively predict a student's unique learning needs and then algorithmically determine the best combination of teaching resources, contents, and methods to meet those needs in real-time?

Methods

The researchers proposed a two-stage optimization pipeline:

  1. Prediction Stage (IACO-SVR):
    • Grey Relation Analysis (GRA): First, they used GRA to identify which "conditional factors" (like interest level, teaching resources, or methods) have the strongest correlation with learning outcomes.
  2. IACO-SVR: They employed Support Vector Regression (SVR) to predict student needs. To avoid the common pitfall of poor parameter selection in SVR, they used an Improved Ant Colony Optimization (IACO) algorithm to find the optimal penalty parameters and kernel functions.
  3. Optimization Stage (IPSO):
    • Once needs are predicted, the system must decide how to teach. The authors used Improved Particle Swarm Optimization (IPSO) to search for the best "teaching decision" combination.
  4. The "Secret Sauce": The IPSO is "improved" by using an adaptive inertia weight and dynamic acceleration factors. This prevents the algorithm from getting stuck in "local optima" (mediocre solutions) and ensures it finds the global best teaching strategy more efficiently than standard PSO.

Datasets

The study utilized a custom dataset collected from an English language classroom context.

  • Data Points: Included student grades, interests, hobbies, and learning progress.
  • Structure: The data was split into a training set (70–80%) and a testing set (20–30%).
  • Categorization: Teaching factors were categorized into seven resource types (e.g., digital platforms, physical models), five content categories (e.g., value-based, skill-based), and nine teaching methods (e.g., flipped classroom, case study).

Key Insights and Ideas

  • Interest is the Engine: Using Grey Relation Analysis, the authors found that Interest Level is the single most important factor affecting learning during the "Preparation" and "Exploration" stages. However, during the "Consolidation" stage, the Teaching Method becomes the dominant factor.
  • The Greenhouse Analogy: The paper interestingly models the classroom like a greenhouse. Just as a greenhouse optimizes light and temperature for crops, this model treats teaching factors as environmental variables that must be tuned to maximize the "yield" of student understanding.
  • Stage-Specific Optimization: The model recognizes that a student's needs aren't static; they shift as a lesson progresses from introduction to deep exploration and finally to review.

Key Results

The proposed IPSO-based model showed significant quantitative improvements over traditional methods:

  • Error Reduction: In the "Preparation Stage," the Mean Absolute Error (MAE) dropped from 1.352 (standard PSO) to 0.625 (IPSO).
  • Model Fit: The determination coefficient ($R^2$) remained high across all stages, peaking at 0.968 during the consolidation stage, indicating the model is highly effective at explaining the variance in student needs.
  • Real-World Impact: In a practical test, student understanding levels (on a 100-point scale) were observed to drop as low as 64 before intervention. After applying the model-driven "teaching decision adjustments," understanding levels rebounded significantly, reaching as high as 99.5.

Limitations and Future Work

While the results are promising, the paper leaves a few doors open for future research:

  • Subject Specificity: The case study focused exclusively on English teaching. It remains to be seen how well this specific multi-conditional factor model translates to highly abstract fields like advanced mathematics or theoretical physics.
  • Data Collection Overhead: The model requires granular data on student "hobbies and interests." In a real-world setting, collecting this data continuously without being intrusive or creating "survey fatigue" is a significant logistical challenge.
  • The Human-in-the-Loop: The paper focuses on the algorithm's output but doesn't extensively discuss how teachers should best receive and act on these AI recommendations in the heat of a live classroom session.

Summary updated

Abstract

With the rapid development of educational technology and the deepening of educational system reform, personalized education has gradually become an important topic in education. However, existing classroom teaching decision-making methods often fail to meet students’ personalized learning needs, resulting in some students being unable to reach their full potential in the classroom. To solve this problem, this study proposed a multi-conditional factor classroom teaching decision optimization method based on the improved particle swarm optimization (IPSO) algorithm, and predicted students’ personalized learning needs by combining with the improved ant colony optimization-support vector regression (IACO-SVR) model. First, the IACO-SVR model was used to collect students’ learning data, such as grades, interests, hobbies and learning progress, to accurately predict their needs in different teaching contexts. Second, the IPSO algorithm was used to optimize the multi-conditional factor classroom teaching decisions, thus meeting the personalized needs of students. The IPSO algorithm had strong global search ability, which effectively found the optimal solution to achieve personalized teaching strategies. It is expected that the teaching quality can be improved by predicting the personalized learning needs of students and optimizing classroom teaching decisions in this study, thus providing better support for their comprehensive development. In addition, the results of this study can provide theoretical basis and reference for administrative departments of education and schools to formulate personalized education policies.
BibTeX
@article{zhang2023personalizing,
  author = {Yi Zhang and Yueyun Shi and Fuli Bi},
  title = {{Personalizing Students' Learning Needs by a Teaching Decision Optimization Method}},
  journal = {International Journal of Emerging Technologies in Learning (iJET)},
  year = {2023},
  doi = {10.3991/ijet.v18i16.42707},
}
Figure