An Empirical Analysis of Data-driven Intelligent Teaching Based on Cloud Class Platform

Xinhong Liu, Yuan Feng, Chunxia Wu, Yaqin Lu, Di GaoAdvances 2020 · DOI

Abstract

At present, with the vigorous development of education platforms, it is necessary to evaluate the effect of teaching processes based on platform data. However, current evaluation methods are often simplistic, lacking data-driven evaluation based on the learning process. This study utilizes a cloud class platform to collect learner data and performs visualization analysis to achieve data-driven learning evaluation. Students are divided into four categories using harmonic curves and cluster analysis, and correspondence analysis is used to identify the strengths and weaknesses of each group. These results guide teachers in mining the relevance between achievements and knowledge points, facilitating individualized teaching and improving overall teaching quality.

Introduction

Despite the rise of digital education platforms, many institutions still rely on "one-size-fits-all" evaluation metrics, such as simple grade averages or GPAs. These traditional methods fail to capture the nuances of the learning process—specifically, where and why a student is struggling. There is a significant gap in data-driven evaluation that tracks the learning process. Without a way to visualize and categorize student performance across specific chapters and knowledge points, teachers cannot effectively implement individualized training strategies. The challenge is to transform raw platform data into actionable insights that allow for "precision teaching"—targeted interventions based on a student's unique academic profile.

Methodology

The researchers employed a multi-stage statistical approach to move beyond simple descriptive statistics:

  • Cluster Analysis (Ward’s Method): Instead of just looking at high or low scores, the authors used the "sum of squares of deviations" to group students. This method minimizes variance within a cluster, ensuring that students in the same group have genuinely similar performance patterns across all tests.
  • Harmonic Curves: This novel visualization technique maps high-dimensional data (multiple test scores) onto a two-dimensional plane as a curve. This allows researchers to visually identify clusters; similar students produce "twisted" or overlapping bundles of curves, making patterns immediately apparent.
  • Correspondence Analysis: This technique was used to find the relationship between student categories and specific knowledge points or test types. It produces a scatter plot where the proximity between a student group and a knowledge point indicates a strong correlation.
Harmonic curve visualization showing the distinct performance patterns of the four student clusters.

Data and Context

The study utilized a real-world dataset collected from a cloud class platform at the Beijing Institute of Petro-chemical Technology, involving 56 undergraduate students. Data points included scores from seven specific tests in "Advanced Mathematics A(I)" and scores from the college entrance examination mathematics (used as a baseline for "Fundamentals"). The tests covered chapters on limits, derivatives, differentials, mean value theorems, and indefinite/definite integrals.

Key Results and Insights

The study successfully classified the 56 students into four distinct categories:

  1. Excellent (5.4%): Balanced performance and high achievers, though they showed relative disadvantages in basic derivatives.
  2. Good (44.6%): The largest group; they possessed the best mathematical foundations but struggled with the application of mean value theorems.
  3. Medium (8.9%): Poor foundations but specific strengths in theorems; they require significant help with integrals and final exam preparation.
  4. Poor (41.1%): Low overall scores with significant "partial chapter" gaps. These students require the most intensive "precision teaching" and self-discipline coaching.
Correspondence analysis plot mapping student categories (blue) to specific knowledge points and test types (red).

One of the most surprising findings was that "Excellent" students were not necessarily those with the best foundational scores. While they excelled in complex areas like definite integrals, they showed relative weaknesses in basic derivatives. Conversely, "Medium" students had the worst foundations but were surprisingly good at the differential mean value theorem, suggesting they need more help with indefinite integrals rather than theory.

Conclusion

This methodology allows teachers to "lock" onto unskilled knowledge points for specific groups, providing a roadmap for targeted feedback and individualized guidance. However, the study was limited to 56 students, and the model needs verification against larger datasets. Future research should incorporate behavioral data—such as platform interaction frequency—to create more holistic learner profiles and test the model's effectiveness across different disciplines.

Summary updated

Abstract

At present, with the vigorous development of education platform, it is necessary to evaluate the effect of teaching process based on the data on the platform, but the evaluation method is relatively simple, lacking the data-driven evaluation based on the learning process. In order to achieve data-driven learning evaluation and improve teaching efficiency, the cloud class platform is used to collect the data of learners' learning process and the analysis and research of data visualization are carried out based on the platform. First, the students are divided into four categories by harmonic curve and cluster analysis. Second, the advantages and disadvantages of all kinds of students are pointed out by correspondence analysis. These results can guide teachers to use education data to mine the relevance between achievements and knowledge points, and help teachers to implement the talent training strategy of individualized teaching, improve teaching quality, provide timely feedback and control for students' learning, facilitate students' independent learning and improve learning effect.
BibTeX
@inproceedings{liu2020empirical,
  author = {Xinhong Liu and Yuan Feng and Chunxia Wu and Yaqin Lu and Di Gao},
  title = {{An Empirical Analysis of Data-driven Intelligent Teaching Based on Cloud Class Platform}},
  booktitle = {Advances},
  year = {2020},
  doi = {10.23977/aetp.2020.41011},
}
Figure