Enhancing Computer Graphics Education through Personalized Recommender Systems with Operation-Based Profiling

Learning Computer Graphics (CG) is a complex process that requires students to master diverse 3D operations within professional software environments. Although recommender systems have been widely studied in education, most applications target domains with abundant structured data, such as mathematics or online courses, and rarely address the unique challenges of CG learning. The absence of domain-specific tools and datasets makes it difficult to provide personalized guidance to students in this field. The present thesis work introduces a novel recommender system embedded directly into Blender, a well-known 3D modeling and animation suite. The system has been designed to diagnose operation-level difficulties and recommend tutorials that address them. Tutorials are represented as documents using the Term Frequency–Inverse Document Frequency (TF-IDF) method, while student profiles are modeled in the same vector space. A Rocchio relevance-feedback algorithm is adapted with an incremental formulation that treats mis- and well-performed operations as feedback events, ensuring rapid adaptation after a single tutorial. Recommendations are generated at the end of each tutorial, while operator-level events are logged continuously for real-time profiling. The system is implemented as an open-source Blender add-on and evaluated through a controlled A/B study. Results show that it unobtrusively builds meaningful profiles after a single tutorial and delivers recommendations aligned with students’ specific difficulties. This work represents the first domain-specific recommender system for CG education and provides a foundation for future hybrid and data-driven approaches.