Using Geometric Features of Drag-and-Drop Trajectories to Understand Students' Learning

Title of the Paper

Using Geometric Features of Drag-and-Drop Trajectories to Understand Students’ Learning

Paper Information

• Authors: Jungwook Rhim, Jiwon Kim, Gahgene Gweon
• Institution: Seoul National University
• Research Domain: Learning behavior analysis and technology-supported education
• Keywords: Drag-and-drop trajectories, decision difficulty, guessing behavior, learning achievement, psychological state
• Conference: CHI 2023, Human-Computer Interaction Conference

Research Background and Problem

• Problem or Challenge:
  • Current educational research lacks non-invasive data analysis methods to understand students' psychological states.
  • The relationship between students' learning behaviors (e.g., guessing behavior and learning achievement) and psychological states remains unclear.
• Importance:
  • Understanding students' psychological states is crucial for optimizing the learning process and enabling targeted interventions by educators.
  • Efficient non-invasive data collection methods can be applied in large-scale learning environments without disrupting students' daily learning activities.
• Research Motivation:
  • Drag-and-drop trajectory data has been used to capture learning-related behaviors, but its comprehensive exploration in relation to psychological states (e.g., decision difficulty) is still insufficient.
  • Proposing a method to infer psychological states and understand students' learning behaviors by analyzing geometric features of drag-and-drop trajectories.
• Related Work:
  • Decision psychology research has demonstrated that trajectory data can reflect users' psychological states.
  • Recently, the education domain has begun to use features such as game logs and response times to detect student behaviors.

Solution

• Proposed Method or Solution:
  • Define a psychological state called "Decision Difficulty Psychological State" (PSD), comprising two subcomponents: conflict and uncertainty.
  • Use eight geometric features of drag-and-drop trajectories (e.g., maximum absolute deviation, mean absolute deviation) to evaluate psychological states.
  • Develop machine learning-based models to predict students' guessing behaviors and learning achievements.
• Innovations:
  • Introduce a refined grouping method for psychological states (PSD), categorizing drag-and-drop trajectory features into "conflict" and "uncertainty."
  • Explore the common associations between psychological states and learning behaviors, validated using a large-scale dataset.
• Implementation Steps and Techniques:
  1. Data Collection: Extract 97,303 trajectory data points from a widely used math game (KitKit School).
  2. Feature Extraction and Grouping: Use exploratory factor analysis (EFA) to categorize geometric features into conflict and uncertainty groups.
  3. Behavior Prediction: Test multiple machine learning algorithms, including linear regression and random forest, to predict guessing behaviors and learning achievements.

Research Results

• Specific Findings:
  1. PSD Composition:
  • Decision difficulty psychological state is divided into two categories: "conflict" (e.g., concern about decision outcomes) and "uncertainty" (e.g., confusion about options).
  • Conflict and uncertainty explain 39.38% and 28.65% of data variance, respectively.
  1. Relationship Between Trajectories and Learning:
  • Conflict positively correlates with the proportion of guessing behaviors (r=0.318, p=0.000) and negatively correlates with learning achievement (r=-0.288, p=0.000).
  • Uncertainty also significantly correlates with the two learning dimensions (guessing behavior proportion r=0.171; learning achievement r=-0.219).
  1. Prediction Model Performance:
  • Using random forest for guessing behavior prediction reduces mean squared error by 35.04%; learning achievement prediction reduces mean squared error by 22.31%.
  • The most important features include distance, mean absolute deviation, and reversal points.
• Advantages Compared to Existing Solutions:
  • Improved interpretability of prediction models by using psychological states to explain abstract geometric features, aiding educators in understanding students.
  • Provides an example of non-invasive data analysis, adaptable to different educational game environments.
• Experiment or Evaluation Results:
  • Offers a data-driven method for assessing psychological states and validates correlations with two common learning behaviors.
  • Prediction models exhibit high confidence and significantly reduced errors.
• Limitations and Future Directions:
  • Data is sourced from a specific age group and country; broader population validation is needed to ensure model generalizability.
  • This study focuses only on two guessing behaviors (quick guessing and systematic guessing); further research could explore other behaviors (e.g., using hints).
  • The method needs adaptation to other interaction forms (e.g., clicks, swipes) to enhance external validity.

Conclusion

This paper elucidates the impact of decision difficulty psychological states on students' learning behaviors and validates the potential of geometric features in predicting learning-related behaviors. Future research can expand the study context and address existing limitations to provide more refined support systems for educational technology development.