Looking for Peer Circles: Graph-Mining-Based Educational Assessment and Refinement Toward Physical Instructors

Exploring the identification of peer circles (or communities) with shared attributes in educational settings provides valuable insights for categorizing instructors’ teaching styles and methodologies. This paper introduces a framework for organizing a large number of physical education instructors into distinct “circles” or clusters based on similar instructional traits and practices. It addresses two main challenges: 1) creating a flexible framework that adapts to diverse instructional characteristics across various educational environments, and 2) handling data insufficiency, as some instructors have limited documented teaching activities. To overcome these challenges, we propose a geometry-based feature selection technique to identify high-quality features that best represent each instructor’s teaching style. An advanced probabilistic model is then applied to represent each instructor’s attributes as a distribution in the latent space. This approach enables the creation of a graph capturing instructional similarities among instructors. By applying an optimized algorithm to detect densely connected subgraphs, we group instructors into circles with shared instructional features. Using these peer circles, we develop a scoring system to evaluate teaching performance. Our experiments, conducted on a dataset of numerous physical education instructors, demonstrate the effectiveness of this approach in identifying distinct teaching styles. Additionally, the framework enhances the assessment and improvement of instructional methods.